Concepts

Distributed-systems concepts: consistency models, CAP, consensus, backpressure, CRDTs, etc.

2903 pages

Most-cited

• Control plane / data plane separation 22 sources — Architectural split between the "decide" path (control plane: validation, authorization, policy, rollout decisions, scheduling) and the "deliver" path (data plane: storage,…
• Observability 21 sources — The function of providing visibility into application performance and reliability via metrics, logs, and traces. The core operational quality it serves: lowering MTTD (mean time…
• Blast radius 19 sources — Blast radius is the scope of damage that a single fault — bug, misconfiguration, vulnerability, runaway workload, compromised credential…
• Compute–storage separation 16 sources — Compute–storage separation is the architectural property where a system's persistence layer and its query/compute layer are decoupled and scale independently.…
• LLM as Judge 15 sources — LLM-as-judge is the evaluation pattern in which one LLM scores another model's (or agent's) output against a rubric — accuracy, helpfulness, policy adherence,…
• Change Data Capture (CDC) 14 sources — Change Data Capture (CDC) is the practice of materialising a table as an ongoing stream of insert / update / delete deltas rather than (or alongside) an authoritative "current…
• Context engineering 12 sources — Context engineering is the discipline of allocating a fixed token budget across the components that compete for the LLM's context window — system prompts, tool descriptions,…
• Open Table Format 12 sources — An open table format (OTF) is a metadata layer over columnar data files on object storage that adds table semantics — atomic row-level updates, schema evolution,…
• Vector Embedding 12 sources — A vector embedding is a dense numerical representation of a piece of unstructured data — text, image, audio, video, or document — produced by an embedding model,…
• Cold Start 11 sources — Cold start names two distinct phenomena that share a name because both describe "the first time around is slow / hard":
• Defense in Depth 11 sources — Defense in depth is the security posture of stacking independent protective layers such that no single compromised or misconfigured layer exposes the whole system.…
• Scale to Zero 11 sources — Scale-to-zero is a service-design property in which an application consumes no capacity and accrues no charge when it has no traffic,…

All pages (A–Z)

• 100 ms slow-query threshold — 100 milliseconds is the threshold PlanetScale uses to flag an individual query as a candidate for optimisation in its query-statistics report.…
• 3D parallelism — 3D parallelism is the standard shape for training billion-plus-parameter language models at multi-node GPU-cluster scale: compose data parallelism (DP), tensor parallelism (TP),…
• 400 VDC rack power — 400 VDC rack power is a data-center power-distribution scheme in which the rack is fed by a 400 V DC bus (as opposed to the more conventional 48 VDC bus-bar used in OCP v1/v2…
• A/B test design audit — A/B test design audit is a quality-review process applied to an A/B test before it goes live, checking that the test design meets agreed-on trustworthiness criteria.…
• Ablation study — forecast type × objective type — An ablation study in ML / optimisation systems isolates the contribution of each design choice by running the full system with one component replaced by a simpler baseline,…
• Abort recovery — Abort recovery is the runtime discipline of detecting a non-unwindable failure — an abort — and reinitialising the sandbox from known-good state so that future operations can…
• Abstract Syntax Tree — An Abstract Syntax Tree (AST) is an intermediate representation (IR) of a program, query, or expression, structured as a tree whose interior nodes are operators / productions…
• Account-per-tenant isolation — Account-per-tenant isolation is the shape of tenant isolation in which the SaaS provider creates a dedicated AWS account per tenant and deploys the tenant's service instances…
• Account takeover (ATO) — Account takeover (ATO) is an incident in which an attacker gains control of an account belonging to a legitimate user or organisation,…
• ACID properties — ACID is a set of four properties — Atomicity, Consistency, Isolation, Durability — that together define the guarantees a database transaction offers.…
• ACK clock — The ACK clock (or self-clocking) is the foundational transport-layer phenomenon by which each round-trip's ACKs from the receiver trigger the sender's next transmission.…
• Producer acks and durability (Kafka) — In Apache Kafka, producers choose per-session how strong a durability guarantee they want per write via the acks property:
• ACL-propagated agent memory — ACL-propagated agent memory is the access-control design where each chunk of agent memory is tagged with its originating data source and the data source's existing ACL is enforced…
• Activation link credential bootstrap — An activation-link credential bootstrap is an onboarding pattern where a service provisions an external user (typically a partner / recipient / tenant) with a one-shot secure URL…
• Active/active replication — Active/active replication is a database-cluster topology with multiple active writers — two or more nodes all accepting writes against the same logical dataset and replicating…
• Active CPU pricing — Active CPU pricing is a serverless-function billing model where customers pay for time the function is actively executing code on-CPU, not wall-clock duration including I/O wait.…
• Active multi-cluster blast radius — Active multi-cluster blast radius is the reliability property obtained by running N independent orchestration clusters active- active,…
• Active-passive failover — Active-passive failover (also: active-standby) is a deployment topology where only one instance (the active) serves traffic at any time,…
• Active/passive replication — Active/passive replication is a database-cluster topology with exactly one active writer (the primary / source / leader) and one or more read-only replicas (the passive nodes /…
• Active/Standby Replication — Active/standby replication is a high-availability topology where one node (active / primary) serves all reads + writes, while a second node (standby / secondary) holds a replica…
• ActiveRecord annotate method — annotate is a method on Rails 7+ ActiveRecord relations that adds a SQL comment to a specific query. It is the per-query override complementing Rails's per-request querylogtags…
• Actor model — The actor model is a concurrency + distributed-systems programming model in which:
• Actor-tagged error — An actor-tagged error is an error occurrence captured in an observability system together with a tag identifying the principal (authenticated user, service account, internal tool,…
• Ad-hoc profiler — Ad-hoc profiler = a user-authored profiler — typically written as a small script in a dedicated DSL — that a profiler platform runs as a first-class profiler without requiring…
• Ad-hoc vs predefined streaming SQL — A structural axis in streaming-SQL substrate design: whether the consumer must declare queries up front so the engine can materialise their results (predefined),…
• Ad impression as hard negative — Ad impressions with no user engagement serve as hard negatives in ads-retrieval contrastive training — served-but-not-engaged items that reflect the real distribution of ads…
• Adapter merging — Adapter merging is the deployment technique of folding the weights of a LoRA adapter (or any low-rank fine-tuning delta) directly into the base model's weight tensors so serving…
• Adaptive bitrate streaming (DASH) — Dynamic Adaptive Streaming over HTTP (DASH) is a streaming technique where the server publishes multiple pre-encoded renditions of the same video — differing in resolution, codec,…
• Adaptive bot reclassification — Adaptive bot reclassification is the bot-management posture in which a session's human-vs-bot label is not final at first classification…
• Adaptive Hash Index — The Adaptive Hash Index (AHI) is an in-memory hash table that MySQL's InnoDB storage engine builds at runtime as an optional overlay on top of its B+tree indexes.…
• Adaptive OOM recovery — Adaptive OOM recovery is the primitive of detecting a task-level out-of-memory error at runtime, restarting the task on a larger VM,…
• Adaptive pagination — Adaptive pagination tunes the page-fetch limit issued to a backing store based on observed item sizes — so that a server expressing a byte-budget SLO to its caller doesn't over-…
• Adaptive paging — Adaptive paging is an alert-routing technique that uses live distributed-tracing causality data at alert time to page the team most likely responsible for the root cause,…
• Adaptive rate limiting — Adaptive rate limiting is a rate-control strategy where the limit (tokens/second, requests/second, or items/second) is periodically recomputed based on current system state…
• Adaptive reasoning effort — Adaptive reasoning effort is the practice of calibrating the depth of an LLM agent's thinking to the complexity of each query,…
• Adaptive Replica Selection (Elasticsearch) — Adaptive Replica Selection (ARS) is Elasticsearch's coordinator-node shard-copy selection algorithm: when a query's shard has multiple copies (primary + replicas),…
• Adaptive upload parallelism — Adaptive upload parallelism is the property of a write pipeline that automatically adjusts the number of concurrent upload streams based on current throughput,…
• Additive Increase Multiplicative Decrease (AIMD) — Additive Increase Multiplicative Decrease (AIMD) is a feedback-control rule for adjusting a transmission-rate variable in response to binary congestion signals.…
• Address-based agent routing — Address-based agent routing is the pattern of encoding the target agent's class and instance identity in the email address itself…
• Adjacency list storage — Adjacency list storage represents a graph by associating each source node with the list of its out-edges (or in-edges) — the direct opposite of an adjacency matrix that would…
• Adversarial review persona — A sub-agent spun up with a prompt / persona that biases it toward critiquing, challenging, and finding flaws in the main agent's output — before any human reviews the work.…
• Advertiser-level loss — Advertiser-level loss is an auxiliary training objective that groups conversion (or other sparse-signal) examples at the advertiser granularity…
• AF_ALG — Linux kernel crypto socket family — AFALG is the Linux kernel's socket-family API for unprivileged user-space access to the kernel crypto API (the same subsystem that handles kTLS, IPsec, dm-crypt, etc.).…
• Agent as first-pass investigator — Agent as first-pass investigator is the operational model in which an AI agent is delegated investigation, diagnosis, and a proposed fix as its bounded scope of authority…
• Agent behavioral baseline — An agent behavioral baseline is a learned characterisation of an AI agent's normal behavioral pattern, derived from historical trace data,…
• Agent brain–hands decoupling — A two-tier architectural shape for managed agent platforms in which the agent loop (model inference, prompt caching, compaction, tool-use orchestration,…
• Agent context window — The agent context window is the fixed-size LLM working set into which an agent session's system prompt, tool descriptions, per-turn messages,…
• Agent-controlled deployment — Agent-controlled deployment is the design posture where an AI agent — not a human release manager — writes the code path, deploys it,…
• Agent D&D alignment framing — The D&D alignment framing is Tyler Akidau's rhetorical device for locating human workers vs. AI agents on a two-axis risk chart adapted from Dungeons & Dragons character…
• Agent-driven browser — Agent-driven browser is an agent-tooling pattern where an LLM agent controls a full browser (DOM + JavaScript state + navigation + event injection + network + console)…
• Agent-driven query fan-out — Agent-driven query fan-out is the workload shape where the consumer of an analytical query substrate is an AI agent that fires hundreds of parallel queries simultaneously…
• Agent-ergonomic CLI — An agent-ergonomic CLI is a command-line interface explicitly designed for AI agents as the primary caller, with human usability as a correlated-but-secondary concern.…
• Agent-first storage primitive — Agent-first storage primitive names the design posture of building a storage service whose API shape, economics, and protocol choices are optimised for the shape of agent workloads…
• Agent hyperfixation failure mode — Agent hyperfixation is the failure mode where an LLM-based coding agent commits to the first hypothesis it generates and works to force that hypothesis to succeed,…
• Agent identity resolution gap — The agent identity resolution gap is the agentic-commerce-specific integration problem that the buyer's identity is not materialised until the buy moment because the agent…
• Agent infrastructure memory — Agent infrastructure memory is a background-populated, long-lived, queryable data store describing the user's runtime infrastructure — services, metric names, deployment topology,…
• Agent loop over a stateless LLM — An LLM is a stateless black box: every call takes an input array and returns an output. The appearance of a multi-turn conversation is an illusion the surrounding program casts…
• Agent memory — Agent memory is an AI agent's accumulated, searchable context across turns and sessions — the things this agent (or this agent for this user) has seen, decided,…
• Agent observability trace tree — Agent observability trace tree extends distributed-tracing concepts (spans, hierarchical parent-child relationships) to LLM agent reasoning loops.…
• Agent orchestration skill — A skill (in the Rovo-Dev / Claude-Code sense of an addressable bundle of agent-facing instructions) that encodes a multi-step workflow, not a single narrow tool binding.…
• Agent payment budget cap — An agent payment budget cap is a default-on ceiling on how much an agent can spend on a given provider in a given period,…
• Agent-provisioned account — An agent-provisioned account is a cloud-provider account that is created on demand by an agent, on behalf of a human user,…
• Agent-provisioned database — An agent-provisioned database is a running, data-bearing database instance that is created on demand by an agent, on behalf of a human user,…
• Agent Readiness Score — Agent Readiness Score is a Lighthouse-style rubric proposed by Cloudflare on 2026-04-17 for grading a website on how well it cooperates with AI agents.…
• Agent self-correction loop — Agent self-correction loop is an intra-trajectory mechanism by which an agent detects its own intermediate-step inconsistencies and revises earlier reasoning without an external…
• Agent Skills discovery — Agent Skills discovery is the pattern of publishing a well-known JSON index at /.well-known/agent-skills/index.json that lists skill documents an agent can ingest to learn how…
• Agent-training-crawler redirect — An agent-training-crawler redirect is the practice of serving AI training crawlers a different URL than humans for the same logical content,…
• Agent wallet over raw credential issuance — An agent wallet is an architectural abstraction in which an AI agent interacts with the user's payment system through a wallet endpoint…
• Agent with root shell — Agent with root shell is a coding-agent deployment posture where the LLM agent's capability surface is a full Linux shell with root privileges,…
• Agentic AI infrastructure challenges (eight-axis checklist) — Tyler Akidau's eight-axis checklist of what an enterprise has to solve to deploy agentic AI safely. Each axis names a concrete infrastructure problem that agent deployment…
• Agentic-commerce bot score — The agentic-commerce bot score is a per-payment classifier output on a checkout flow that estimates the likelihood the payment was made by a malicious bot rather than…
• Agentic commerce — Agentic commerce is the class of buy transactions where an AI agent acts as the proxy for a human buyer — surfacing products, verifying inventory, completing checkout,…
• Agentic Data Access — Agentic data access is the design concern that as applications are written increasingly by AI agents, the cost of friction between agent and data compounds into reasoning overhead…
• Agentic data classification — Agentic data classification is the use of multi-signal classifiers — combining pattern recognition, schema/metadata features,…
• Agentic development loop — The agentic development loop is a closed-loop LLM code generation workflow: the LLM proposes code, an execution environment runs the code, the environment's output (stdout,…
• Agentic development — Agentic development is a development model where an AI agent does more than suggest snippets — it "writes, tests, deploys,…
• Agentic Kernel Synthesis — Agentic kernel synthesis is the system-level framing that couples four primitives into a production-grade pipeline for authoring GPU/accelerator kernels without hand-tuning:
• Agentic Paywall — Agentic paywall is the design posture where content access is priced per request and an AI agent, given a spending budget by its human principal,…
• Agentic traffic share — Agentic traffic share is the fraction of HTTP traffic originating from AI-driven automated clients — LLM-based agents, AI-assisted development tools,…
• Agentic troubleshooting loop — An agentic troubleshooting loop is an investigation pattern in which an LLM is the planner and a narrow-surface tool assistant is the hands, iterating through a cycle of:
• Agentic workflow governance — Agentic workflow governance is the discipline of applying security, compliance, and operational policy at the level of what autonomous AI agents do — their tool invocations,…
• Aggregate demand smoothing (multi-tenant scale) — Aggregate demand smoothing names the observation that, in a system shared by a very large number of independent tenants, the sum of bursty per-tenant workloads is much smoother…
• Aggregate tag attribution — The capability to attribute aggregate query statistics (total runtime, row counts, execution count, cumulative resource consumption) to individual tag values…
• Aggregation pipeline — Aggregation pipeline is MongoDB's declarative server-side query framework: an ordered sequence of stages ($match, $project, $group, $addFields, $lookup, $sort, $limit, $unwind,…
• AI agent guardrails — AI agent guardrails is the discipline of running AI-generated code through the same (or stronger) quality gates that human-written code would face,…
• AI-assisted codebase rewrite — An AI-assisted codebase rewrite is a project where one (or few) humans direct a capable AI coding agent to produce the overwhelming majority of code for a non-trivial target…
• AI-assisted refactoring economics — The emergence of capable AI coding assistants fundamentally alters the cost-benefit analysis of addressing technical debt.…
• AI instructions as code — When writing documentation for AI agent consumption, treat the instructions themselves as code: versioned, reviewed, iterated, formally structured,…
• AI-provenance UI indicator — An AI-provenance UI indicator is a visual marker attached to a value in a UI whose sole purpose is to disclose that the value was suggested by an AI system rather than entered…
• AI re-review (incremental) — Incremental AI re-review is the discipline of re-running an AI code-review agent with full awareness of its own prior findings and the human responses to them,…
• AI thinking heartbeat — AI thinking heartbeat is the operational discipline of emitting a periodic "the model is still working, here is how long it's been since any output" log line while a long-running…
• AI writes own tests — The AI agent writes both the production code and the test suite that validates it — the agent produces the e2e tests, deploys to a real environment, runs the tests,…
• Airflow DAG zip packaging — Packaging DAGs is an Airflow feature that deploys a DAG and all of its Python dependencies as a single zip archive dropped in Airflow's dags/ folder.…
• Airflow TaskGroup Parallelism — Airflow TaskGroup parallelism is the DAG-structuring discipline of placing N instances of the same logical pipeline into N separate TaskGroup subgraphs inside one DAG,…
• Aleatoric uncertainty — Aleatoric uncertainty is the component of prediction uncertainty that comes from inherent randomness in the system being modelled — not from limited training data,…
• Alert fatigue — The operator-side failure mode in which a notification channel emits so many alerts — especially low-signal / duplicate / flapping…
• Always be failing over — Always be failing over is the reliability-process principle that failover is exercised routinely, not reserved for emergencies.…
• Always-on ambient sensing — Always-on ambient sensing is the serving-infra envelope where an ML model runs continuously, on a small battery, inside a body-worn or environment-deployed device,…
• Amazon Time Sync attribution — Amazon Time Sync is AWS's fleet-wide time-synchronisation service, exposed as a local NTP/PTP endpoint on Nitro-based EC2 instances.…
• Analytical intent embedding — An analytical intent embedding is the vector representation of "the business question a historical SQL query was designed to answer" — as opposed to the raw SQL, the table schema,…
• Analytical question bridge — The analytical question bridge is the load-bearing design trick in Pinterest's Text-to-SQL system: the SQL-to-text transformation step produces an explicit list of "analytical…
• ANN (approximate nearest neighbor) index — An ANN (approximate nearest neighbor) index is a data structure + serving system that, given a query vector, returns the top-K closest item vectors from a large pre-indexed corpus…
• Anomaly-gated config update — An anomaly-gated config update is a deployment discipline for learned or computed configuration artefacts in which each new version is compared against the previously-published…
• Anomaly vs incident separation — Anomaly vs incident separation is an incident-management process pattern in which alerts that reach on-call engineers are triaged into two distinct tracks:
• Anonymous credential — An anonymous credential is a cryptographic primitive that lets a client prove the result of a check (e.g. "I solved a challenge", "I have good standing with this service",…
• Anti-entropy repair pause — Pausing anti-entropy repairs during a rolling upgrade is a pre-flight discipline specific to datastores with an explicit repair / reconciliation pass (most prominently Cassandra's…
• Anti-entropy (gossip) — Anti-entropy is the family of gossip algorithms in which nodes compare full replicas (or their compact summaries) and patch the differences. Coined by Demers et al.…
• Anti-flapping — Anti-flapping is the operational discipline of rate-limiting leadership changes in a consensus system: after a leadership change completes,…
• Anycast — Anycast is a network-layer routing discipline in which the same IP address (or prefix) is advertised from multiple geographically dispersed points of presence (POPs) via BGP,…
• Anytime optimization algorithm — An anytime algorithm is an optimization procedure that produces a valid-but-possibly-suboptimal answer at every step, and improves that answer monotonically with more iterations.…
• API Catalog — API Catalog is an IETF standard (RFC 9727, 2025) that defines a well-known URL at which a service publishes a JSON document listing all of its public APIs: links to specs (OpenAPI…
• API-first principle — API-first is the discipline of defining a service's external API contract (typically as an OpenAPI / Swagger document) before writing the implementation,…
• API normalization for multi-cloud LLM serving — API normalization for multi-cloud LLM serving is the design discipline of providing a single internal contract to feature teams that hides per-provider differences in API shape,…
• Appcraft interoperability — Interoperability in Appcraft's sense is the problem of mixing Appcraft-rendered content with non-Appcraft (classically-native) content inside a single mobile app session:
• Appropriate Technology — Appropriate technology is Werner Vogels' framing for choosing the stack that fits the customer's actual constraints, not the stack that is most advanced.…
• Approximate top-K sketch — An approximate top-K sketch is a streaming probabilistic data structure that tracks the K most-frequent items in a stream of events,…
• Approximate vs exact kNN — The Elasticsearch-named distinction between two families of k-nearest-neighbor search over vector embeddings:
• Architectural defense vs faster patching — When AI-driven vulnerability discovery shortens the attacker's timeline (CVE → working exploit → fleet exploitation), security teams have two structurally different responses…
• Architectural fitness function — An architectural fitness function is a machine-executable test that asserts an architectural rule about a codebase, run continuously in CI to catch deviations at the moment they…
• Architectural Isolation — Architectural isolation is the multi-tenant security posture where tenants do not share the plane on which unencrypted data and credentials live…
• Architecture over patching — Architecture over patching is the principle that the architecture around a vulnerability determines the damage of exploitation more than the speed at which the vulnerability…
• Artifact-as-API — Artifact-as-API is the design principle that inter-agent (or inter-role) communication should happen exclusively through schema-validated files on disk rather than through…
• AS-path prepending — AS-path prepending is a BGP traffic-engineering technique in which an AS pads its own ASN into the outbound AS path more than once before advertising a route,…
• ASN rotation — An autonomous system number (ASN) identifies a network under a single administrative policy for BGP routing. ASN rotation is the practice of sourcing request traffic from multiple…
• Asset-backed bet — A startup-economics framing: bets that acquire hardware or other tradable assets with durable resale value have a fundamentally different risk profile than bets whose cost is pure…
• Asset-first agent design — Asset-first agent design is the principle that an LLM agent should surface existing, trusted assets — curated tables, queries, dashboards,…
• Assume-compromise posture — Assume-compromise posture is the security-team discipline of, on any critical vulnerability disclosure, starting from the assumption that compromise has already occurred…
• AST interpreter — An AST interpreter (also tree-walking interpreter) is the simplest strategy for executing a dynamic language or expression language: parse the input into an abstract-syntax-tree,…
• AST literal normalization — AST literal normalization is the technique of producing a canonical form of a SQL statement by parsing it to an abstract syntax tree, walking the tree,…
• Asymmetric attacker-defender gap — The asymmetric attacker-defender gap is the structural inequality in security: an attacker only needs to find and exploit one opening,…
• Async agent workflow — Async agent workflow is a development-iteration shape where agents pick work from a queue (issues, tickets, tasks) and surface results as asynchronous artifacts (PRs,…
• Async cancellation thread-spawn anti-pattern — The async cancellation thread-spawn anti-pattern is the failure mode where a component, faced with the need to cancel in-flight work in a way that would deadlock if done…
• Async clone + background hydration — Async clone + background hydration is the repository-materialisation shape where a clone-equivalent operation returns as soon as the file tree + refs are present and proceeds…
• Async context propagation — Async context propagation is the mechanism by which a distributed-tracing library associates a span (or other per-request context) with the asynchronous continuation of the code…
• Async CPU-GPU pipelined scheduling — Async CPU-GPU pipelined scheduling is the inference-scheduler discipline of moving CPU-side post-processing of batch N off the critical path so it runs concurrently with the next…
• Async I/O concurrency threshold — Async I/O concurrency threshold is the observation that asynchronous-I/O interfaces (like Linux's iouring) only outperform synchronous I/O above a certain concurrency / I/O-rate…
• Async iteration — Async iteration is the JavaScript language protocol for consuming asynchronous sequences: an object implementing Symbol.asyncIterator + next() returning a Promise<{ value,…
• Async Kafka publication for telemetry — Async Kafka publication for telemetry is the pattern where a hot-path producer (query serving layer, RPC handler, application thread) writes telemetry to a bounded in-memory buffer…
• Async-loading cache stale window — An async-loading cache stale window is a trailing segment of an entry's TTL during which reads are served from the cached value immediately while a background refresh against…
• Async preemption (Go) — Async preemption is the Go runtime's mechanism — introduced in Go 1.14 — for forcibly pausing a goroutine that has been running too long,…
• Async relationship inference — Async relationship inference is the metadata-graph primitive of deriving cross-system relationships in scheduled background jobs,…
• Async Rust Future — A Future in async Rust (std::future::Future) is a lazy state machine representing the eventual value of an asynchronous computation.…
• Async write buffer — An async write buffer is an in-process buffer (a channel, queue, or ring) sitting between a synchronous request path (e.g. an HTTP handler) and a downstream durable write (e.g.…
• Asynchronous cascade delete — Asynchronous cascade delete is a strategy for removing a node (or other parent record) from a high-fanout graph store where:
• Asynchronous replication — Asynchronous replication is a replication posture in which the primary (source) system acknowledges a write to the client before secondary (replica) systems have applied it.…
• Asynchronous reply email — Asynchronous reply email is the property of the email channel that permits a reply to arrive an arbitrary amount of time after the inbound message — seconds, minutes, hours,…
• Asynchronous sequence packing — Asynchronous sequence packing is a dataloader pattern for variable-length LLM training: pack multiple short samples into fixed-length sequences with a document mask that prevents…
• AsyncRead contract — AsyncRead (and its sibling AsyncWrite) are async Rust traits that extend Future to streaming I/O. Where a one-shot Future resolves once and then can't be polled again,…
• At-least-once delivery — At-least-once delivery is the messaging guarantee that each message will eventually be delivered to a consumer at least once,…
• At-least-once uploads for cost reduction — In a 3-way-replicated stateful system that tiers data out to object storage, the naive approach is to have each replica upload its local blocks to object storage…
• At-most-once delivery — At-most-once delivery is a messaging / RPC delivery semantic in which each message is delivered zero or one times, never more.…
• Athena shared-resource contention — Amazon Athena (amazon-athena) is a shared serverless query engine: queries run on a pooled Presto/Trino cluster whose capacity is allocated across all AWS customers in the region.…
• Atomic distributed transaction — An atomic distributed transaction is a multi-participant write protocol (typically 2PC or a 2PC variant) that provides all-or-nothing atomicity across multiple database…
• Atomic product ID vs Semantic ID — A vocabulary-substrate trade-off in recsys retrieval: should the recommender's tokens be atomic product IDs (one token per item, opaque) or Semantic IDs (short codeword sequences,…
• Attack surface minimization — Attack surface minimization is the design discipline of keeping the set of code paths / APIs / parsers / features reachable by untrusted input as small as possible.…
• Attention-Based Fusion (Multimodal) — Attention-based fusion is the multimodal-ML strategy of learning dynamic per-example weights across modalities and over time,…
• Attenuation (offline) — Attenuation is the operation of narrowing a Macaroon's privileges by appending additional caveats — restrictions — to the token.…
• Attribute-based access control (ABAC) — Attribute-based access control (ABAC) decides whether a principal may perform an action on a resource by evaluating a policy over attributes of all three (plus context)…
• Audio codec — An audio codec (coder-decoder) compresses raw audio for transmission or storage and reconstructs it at the receiver. For real-time communication (RTC),…
• Audit log as transparency artifact — An audit log as transparency artifact is an append-only record of every high-stakes operation a trust authority performs,…
• Audit trail in OpenSearch — Keeping a permanently-retained audit log of security- sensitive operations in a long-lived search cluster (Fly.io uses OpenSearch) as a load-bearing architectural layer…
• Audit trail — An audit trail is a durable, queryable record of every state-changing operation performed against a system — who changed what, when,…
• Augmented Inverse Probability Weighting (AIPW) — AIPW is a doubly-robust observational causal estimator that combines two nuisance models:
• Aurora storage quorum — Aurora storage quorum is the replication scheme that Amazon Aurora uses internally within a single cluster to make writes durable: redo-log entries produced by the writer compute…
• Authoritative vs fast notification — A two-tier correctness posture for real-time applications in which one channel carries durable, authoritative state changes (typically a transactional database commit)…
• Authorization as a service — Authorization as a service is authorization delivered as a platform capability: application teams opt in with a single declarative statement (e.g.,…
• Authorization decision caching — Authorization decision caching stores the Allow / Deny outcome of an authorization check for later reuse, so subsequent requests with the same inputs skip the policy-engine round…
• Authorization vs authentication token — Separating authorization (what the bearer may do) from authentication (who the bearer is) into two distinct tokens — instead of conflating them in one…
• Auto-regressive grain model — An auto-regressive (AR) grain model is a compact parametric generator for noise patterns used by AV1 Film Grain Synthesis.…
• Auto-save modal UX coherence — Auto-save modal UX coherence is the discipline of making settings-modal persistence behavior match the architectural premise of the settings it exposes.…
• Autoformalization — Autoformalization is the problem of translating natural-language specifications (policies, procedures, guidelines, manuals, informal rules) into formal,…
• Automated backup validation — Automated backup validation is the property that every backup produced by a database platform is proven to be restorable as part of the backup creation pipeline…
• Automated circuit breaker with partial-open recovery state — The automated circuit breaker with partial-open recovery state is a refinement of the classical circuit breaker specifically tuned for LLM-serving routing layers.…
• Automated reasoning — Automated reasoning is the family of techniques that mechanically prove (or disprove) properties of systems against a formal specification — SAT solvers, SMT solvers,…
• Automated root-cause analysis — Automated root-cause analysis (RCA) is the practice of encoding enough operational knowledge in tooling that, when an alert fires,…
• Automated Test Generation from Production Traffic — Automated test generation from production traffic is the discipline of producing a representative test set for an offline evaluation by sampling from production request / event…
• Automated vs manual testing complementarity — Automated vs manual testing complementarity is the framing that automated testing is a layer in a broader testing strategy, not a substitute for manual testing.…
• Automatic dependency tracking — Automatic dependency tracking is the mechanism where a reactive framework records the dependency graph implicitly, by observing what each computation reads at runtime,…
• Automatic row limit — An automatic row limit is a Vitess query-serving safety feature that caps the maximum number of rows a single query can return at the tablet level,…
• Automatic table optimization — Automatic table optimization is the substrate property where the data platform — not the user — owns scheduling and execution of compaction (OPTIMIZE), retention cleanup (VACUUM),…
• Automatic task caching — Automatic task caching in workflow orchestration detects when a pipeline task has already been executed with identical parameters and inputs,…
• Autonomous attack agent — An autonomous attack agent is an LLM-driven automated agent that systematically probes CI/CD systems, code repositories, or other attack surfaces across many targets in parallel,…
• Autonomous System (AS) — An Autonomous System is an administratively-unified network that speaks BGP with its neighbors under a single routing policy.…
• Autonomy (enterprise agents) — Autonomy is the shift from prompt-response one-shot LLM calls to code-in-control end-to-end flow, where the LLM is empowered to make function calls, choose which database to read,…
• Auxiliary-task regularization — Auxiliary-task regularisation is a multi-task-learning technique in which one or more auxiliary tasks — often abundant, cheap,…
• AV1 layered coding — AV1 layered coding is a main-profile AV1 feature that lets a single bitstream carry multiple layers of video content, where each layer is independently decodable and can be added…
• Availability Dependency — An availability dependency exists when service A cannot serve its SLO without service B being available. A's availability is upper-bounded by B's: if B is down, A is down.
• Availability multiplication of dependencies — A service synchronously depending on N downstream components has a request availability upper-bounded by the product of each component's individual availability.…
• Availability vs data-loss trade-off — The availability-vs-data-loss trade-off is the operational decision a consensus-backed system must make when concurrent failures exceed the failure-tolerance envelope: either stall…
• Availability Zone balance — Availability Zone (AZ) balance is the property of a workload whose replicas are evenly distributed across the cloud provider's AZs,…
• Availability zone failure drill — An availability zone failure drill is a chaos-engineering exercise that simulates the complete loss of a single cloud availability zone and verifies the fleet re-balances…
• AWS partition — An AWS partition is a logically isolated group of AWS Regions with its own set of resources, including IAM. Partitions exist to meet country- or use-case-specific compliance,…
• AZ-bucketed environment split — AZ-bucketed environment split is the engineering choice of partitioning a single fleet-configuration environment (e.g., a Chef prod environment) into N parallel environments keyed…
• B+tree — A B+tree is a B-tree variant with two structural modifications that make it dominate database indexes over the plain B-tree:
• B-tree — A B-tree of order K is a self-balancing tree where each node stores between 1 and K key/value pairs (internal nodes ≥K/2), every node has N+1 children for N keys,…
• Back-dirty (feedback-loop invalidation) — A back-dirty is a feedback-loop pattern in a multi-system reactive runtime where a later-running system invalidates work done by an earlier-running system,…
• Back-of-the-envelope estimation — Back-of-the-envelope estimation (BOTE) is the engineering discipline of arriving at a rough-order-of-magnitude sizing number for a proposed system — capacity, cost, latency,…
• Backend-for-Frontend (BFF) — Backend-for-Frontend (BFF) is an architectural pattern pioneered at SoundCloud (sam-newman.com) where a backend service is created per frontend surface or per business/use-case…
• Backend/frontend network separation — In large GPU training clusters, backend/frontend (BE/FE) network separation is the practice of connecting each training rack to two distinct physical networks with different design…
• Background hydration — A cache-warming / replica-fill technique in which a system serves reads against a remote authoritative source while a background process downloads the full data set to local…
• Backpressure — Backpressure is the control-plane primitive by which a slow consumer in a streaming pipeline signals a fast producer to slow down.…
• Backup encryption at rest — Backup encryption at rest is the property that backup artifacts stored in object storage (S3, GCS, Azure Blob) are stored in encrypted form,…
• Backward compatibility in data models — An explicit design requirement that introducing new product data into a data warehouse must not break existing models that power the core product line's reporting and analytics.
• Backward compatibility — Backward compatibility is the property that interfaces accept and behave correctly on inputs + requests that were valid under prior versions of the interface.…
• Bad-host detection — Bad-host detection is the discipline of identifying individual hosts in a fleet that are functioning badly without being unresponsive — they pass cluster-level health checks,…
• Bandwidth per terabyte (BW/TB) — Bandwidth per terabyte (BW/TB) is the ratio of a drive's sustainable read/write throughput (in MB/s) to its usable capacity (in TB).…
• Base-class instrumentation — Base-class instrumentation is the practice of placing cross-cutting measurement / logging / analytics logic into an ancestor class that all relevant subclasses inherit from…
• Base-image unused-systemd-unit risk — Cloud base images (AMIs, Docker base images, distro golden images) typically ship with pre-configured systemd units active by default because the image vendor has to serve the most…
• Base64 as byte buffer (avoid string materialisation) — Base64 as byte buffer is the performance technique of reading Base64-encoded data directly from its file-buffer representation…
• Batch-only component for IPC amortization — Batch-only component for IPC amortization is the design choice to expose only the batch variant (BatchInput / BatchProcessor / BatchOutput) of a component type across…
• Batching latency trade-off — The batching latency trade-off is the explicit exchange a producer makes when it groups records into batches: higher throughput is paid for with higher per-record latency.…
• Bayesian optimization over parameter space — Bayesian optimization is a class of sequential model-based optimization algorithms for expensive-to-evaluate objective functions.…
• BBR congestion control — BBR (Bottleneck Bandwidth and Round-trip propagation time) is a model-based congestion-control algorithm introduced by Google in 2016 (ACM Queue article).…
• Beam search retrieval — Beam search retrieval is the use of beam search at inference time to generate K distinct candidate item identifiers from an autoregressive recsys decoder.…
• Vulnerability-agnostic behavioral detection — Vulnerability-agnostic behavioral detection is an endpoint- detection discipline in which the detection engine fires on anomalous process-execution patterns…
• Benchmark methodology bias — A benchmark methodology bias is a confounder built into a benchmark's setup (not its subject) that systematically skews results in one direction — and critically,…
• Benchmark representativeness — Benchmark representativeness is the measurable property that a benchmark's execution behaviour matches the distribution of execution behaviour of the target workload.…
• Best-effort log delivery — Best-effort log delivery names the reliability contract on the loosest end of the streaming-delivery spectrum: a log record may be missed, arrive late, or be duplicated,…
• Best-Effort vs. Eventually Consistent counter — A distributed counter service can expose its counting use cases as one of two primary modes with explicit trade-offs:
• BF16 exponent redundancy — BF16 exponent redundancy is the empirical observation that the 8-bit exponent byte of BF16 weights in trained LLMs is distributed extremely non-uniformly: out of 256 possible…
• BGP community — A BGP community is a 32-bit (or 64-bit for large communities) tag attached to a route that flows alongside the AS path in BGP advertisements.…
• BGP export policy — A BGP export policy is the set of rules that decides which routes an AS will advertise to a given neighbor. It is the per-AS,…
• BGP route hijack — A BGP route hijack occurs when an attacker advertises BGP routes for IP prefixes they are not authorized to originate, redirecting traffic away from the legitimate destination.…
• BGP route withdrawal — BGP route withdrawal is the act of unadvertising a prefix from a BGP speaker — once withdrawn, transit and peer routers remove the prefix from their tables and stop forwarding…
• Border Gateway Protocol (BGP) — BGP is the path-vector routing protocol that glues the Internet together. Each Autonomous System (AS) speaks BGP with its neighbors to exchange reachability information…
• Bidirectional RPC — Bidirectional RPC is an RPC model in which the protocol is symmetric — there is no privileged "client" and "server" at the protocol layer.…
• Big-O over key length — Standard algorithmic analysis talks about Big-O over the number of stored elements n — a hashmap read is O(1) whereas a sorted-tree lookup is O(log n).…
• Bin-packing — Bin-packing is the combinatorial-optimisation problem of placing a set of items with known sizes into the smallest number of fixed-capacity bins (or equivalently: maximising…
• Binary authorization — Binary authorization is the security control that restricts a device or host to running only pre-approved executable binaries. Everything else is blocked at exec-syscall time.…
• Binary-size bloat — Binary-size bloat is the monotonic growth of a compiled artifact over time as features + dependencies accumulate, without a commensurate removal discipline.…
• Binary vs. graded LLM scoring — Binary scoring uses True/False per-criterion verdicts from an LLM judge; session-level or item-level holistic scores aggregate many binary verdicts.…
• Binlog replication — Binlog replication is MySQL's native change-data-capture mechanism: every committed transaction is appended, in order, to the server's binary log (binlog),…
• Bisection bandwidth — Bisection bandwidth is the aggregate bandwidth available across the narrowest cut that divides a network in half. For a non-blocking Clos fabric,…
• Bisimulation — Bisimulation is the two-way equality of behaviour-sets between two state machines (typically a specification and an implementation): every behaviour of one is realizable…
• Bitpacking — Bitpacking is the practice of combining multiple sub-byte elements into a native machine data type (uint8, int32, …) for storage and transport,…
• Bitrate ladder — A bitrate ladder is the set of pre-encoded renditions of the same source content that a streaming service publishes for adaptive bitrate (ABR) playback.…
• Bitwise double-free — A bitwise double-free is the race-condition shape where two code paths attempt to atomically clear the same bit of a tightly-packed state word without coordinating…
• Black-box validation — Validate a system by its observable inputs and outputs, not by reading its implementation. Submit a known input, check the output against expected behaviour,…
• Blast-radius multiplier at fleet scale — Blast-radius multiplier at fleet scale is the statistical reality that, as the number of components in a fleet grows, the probability of at least one active impacting event tends…
• Blast radius — Blast radius is the scope of damage that a single fault — bug, misconfiguration, vulnerability, runaway workload, compromised credential…
• BLOB/TEXT index prefix requirement — MySQL requires that any B-tree index on a BLOB or TEXT column declare a prefix length — the number of leading bytes the index covers…
• Bloblang — Bloblang is Redpanda Connect's built-in declarative mapping, filtering, and transformation language — a small DSL for per-record reshape operations inside a Redpanda Connect…
• Block-level async clone — Block-level async clone is the storage-migration primitive where a source block device is cloned asynchronously into a destination device: the clone is immediately usable (attaches…
• Block verification — Block verification is a speculative-decoding verification primitive in which the expert model accepts or rejects a block of N drafted tokens jointly,…
• Bloom filter — A Bloom filter is a space-efficient probabilistic data structure that tests whether an element is a member of a set. It returns:
• Bloom index filter (GPU-native eligibility filter) — The Bloom index filter is the GPU-native eligibility-filtering primitive used inside Meta's SilverTorch retrieval substrate (Source:…
• Blue/green deployment — Blue/green deployment is a deployment strategy where two identical production-capable environments (named "blue" and "green") are maintained,…
• Bolt-vs-SQLite storage choice — The design decision between a key-value embedded store (e.g. BoltDB) and a full-SQL embedded store (e.g. SQLite) for a service's state.…
• Boolean query DSL — A boolean query DSL is a structured query-document shape in which clauses combine via explicit AND / OR / NOT operators — usually rendered as nested tree objects…
• Bootstrap percentile method — The bootstrap percentile method is a non-parametric technique for computing confidence intervals (and thus significance) for a point estimate. Given a sample of size n:
• Bootstrapping circular dependency — A bootstrapping circular dependency is the variant of circular-dependency that manifests during cold-start recovery of a system whose control plane depends on itself to function.…
• Bot-safer-than-human — Bot-safer-than-human is the deliberate decision to automate transformations that are not strictly necessary, precisely because they are delicate enough that human reviewers will…
• Bot-vs-human frame — The bot-vs-human frame is the assumption that the important question at the origin is "is this request from a person or a program?".…
• Boundary as Feature — A design principle crystallised in the 2026 S3 Files post: when two abstractions (e.g. file and object, stream and batch,…
• Bounded fan-out relevance cap — Bounded fan-out relevance cap is the discipline of capping the worst-case set that a scatter-gather fallback fans out over…
• Branch-level cost attribution — Branch-level cost attribution is the property that compute and storage cost line items in the platform's billing substrate are broken down automatically by branch identity…
• Branch-level governance propagation — Branch-level governance propagation is the property that governance policies (row filters, column masks, attribute-based access controls) attached to a database's catalog scope…
• Break-glass escape hatch — Break-glass escape hatch is an explicit, telemetry-tracked mechanism for a human to override an automated gate when the automation is wrong, outaged,…
• Breaking-change detection via static analysis — A use-case framing for fitness-function-style static analysis: instead of "deprecate-and-pray" — mark an API @Deprecated, hope nobody depends on it,…
• Breaking change requires major bump — A breaking change to a component (or API, or message schema) requires a major version bump. Once the major version has bumped,…
• Broadcast FACS/FAX pre-show check — FACS/FAX ("facilities checks") is the pre-show rehearsal gate before every live broadcast — a battery of Audio/Video tests run during rehearsal that catches venue-side issues…
• Broadcast operator role specialization — The decomposition of a "broadcast operator" into distinct specialised roles — separating inbound signal handling from outbound stream handling from qualitative signal QC…
• Broker-side effective-batch-size observability — Broker-side effective-batch-size observability is the practice of measuring the effective batch size arriving at a streaming broker (as opposed to the producer's configured…
• Broker-internal cross-cluster replication — Broker-internal cross-cluster replication is the architectural property where the broker itself implements cross-cluster replication…
• Broker partition density — Broker partition density is the supported partition count per streaming-broker cluster tier — a first-order capacity-planning parameter that maps a customer's tier-sizing decision…
• Broker-side write caching — Broker-side write caching is a broker configuration that relaxes the produce-ack latency path by acknowledging writes once the record is in memory on a quorum of brokers,…
• Brownfield RN integration — Brownfield RN integration is the problem of embedding React Native into a large, mature, already-shipping native iOS/Android codebase — as opposed to greenfield RN,…
• Browser telemetry fingerprint — A browser telemetry fingerprint is a composite signature computed from client-side-collected properties of a running browser session: canvas / WebGL / audio-context rendering…
• BSON document overhead — BSON document overhead is the per-document + per-field bytes that MongoDB's on-the-wire / on-disk BSON binary encoding pays regardless of user data: a document length prefix,…
• Bucket join — A bucket join is a distributed join executed over tables whose rows are pre-partitioned and sorted by the join key into matching buckets (file groups, partitions).…
• Bucket Pattern — Bucket Pattern — MongoDB's named schema-design pattern where many fine-grained events (one event = one would-be document) are grouped into a single bucket document by a shared key…
• Buddy allocator shard grouping — Buddy allocator shard grouping is the adaptation of the classic OS buddy-allocator algorithm (power-of-two memory block splitting/merging) to the problem of dynamically adjusting…
• Build-error-driven fix loop — A build-error-driven fix loop is a migration-pipeline phase that treats the compiler's error messages as the specification for its next edits: run the compiler, parse the errors,…
• Build graph — A build graph is the DAG of declared build/test actions and their dependencies. Nodes are actions (compile-this, link-that, run-this-test);…
• Build parallelism for ingest serialization — Build parallelism is the thread-count knob for the serialisation-and-commit-preparation step in an analytical-store streaming ingest connector.…
• Build reliability — Build reliability is the developer-facing metric for how often a build (CI pipeline on the main / target branch) succeeds.…
• Build time scales with page count — When a static-site / SSG tool pre-renders every page listed in the enumerator (Next.js's generateStaticParams(), Gatsby's createPages, etc.) during the build,…
• Build-time vs runtime phase separation — Build-time vs runtime phase separation is the decomposition axis that splits a framework-integration problem into two temporal layers:
• Build-vs-buy for CMS — Build-vs-buy for CMS is the architectural decision a platform-engineering team makes when it needs editorial content-management capability for its product surfaces…
• Bump-in-the-wire middlebox — A bump-in-the-wire middlebox is a network device — firewall, IDS/IPS, DLP, encryption/decryption, transparent proxy — that is inserted into a traffic path by routing,…
• Bundled GitOps dependency anti-pattern — Bundling a GitOps controller (e.g. FluxCD, ArgoCD) as a hidden internal dependency of a Kubernetes Operator or similar vendor-shipped platform component is a structural…
• Bundler chunk invalidation — Bundler chunk invalidation is the phenomenon where a small source-code change causes a JavaScript bundler (webpack, Vite, Rollup, esbuild, Turbopack,…
• Bursty query pattern — Bursty query pattern is a workload shape where reads concentrate into short, high-intensity bursts rather than flowing at a steady rate.…
• Bus-hop storage tradeoff — The bus-hop storage tradeoff is the architectural bargain a cloud operator makes when deciding whether a compute host's storage lives on the same physical (local NVMe,…
• Business-group authorization gating — Business-group authorization gating is the access-control pattern of restricting a service's effective user population at session-establishment time by checking a business-group…
• Butterfly effect in complex systems — The butterfly effect names the property of complex systems that small inputs can produce disproportionately large outputs…
• BYO workflow per tenant — BYO workflow per tenant — bring-your-own-workflow-per-tenant — is the platform shape where the customer authors the durable run(event,…
• BYOB reads (bring-your-own-buffer) — BYOB (bring-your-own-buffer) reads are the Web Streams mechanism for reducing per-chunk allocation in byte-oriented streams: instead of the stream allocating a new buffer for each…
• BYOC data ownership for Iceberg — BYOC data ownership for Iceberg is the property that falls out when Redpanda BYOC (data plane in the customer's cloud account) is combined with Iceberg Topics (broker writes…
• BYOK (Bring Your Own Key) — BYOK (Bring-Your-Own-Key) is the posture in which a customer stores a third-party provider API key (OpenAI, Anthropic, Google,…
• Byte-size pagination — Byte-size pagination is the API-design choice of bounding a paged response by total byte size rather than by number of items.…
• Bytecode virtual machine — A bytecode virtual machine (bytecode VM) is a dynamic-language execution strategy that sits between AST interpreters and JIT compilers on the complexity/performance spectrum:
• Bytecode vs AST static analysis — A structural choice in JVM static-analysis tool design: whether to operate on the Abstract Syntax Tree (AST) of source code (via a parser) or on the compiled bytecode (via…
• Bytes in flight — Bytes in flight (abbreviated bytesinflight, sometimes bif) is the number of bytes the sender has transmitted but not yet acknowledged at any moment.…
• C++ compilation model (pre-processing & transitive includes) — C++ compilation happens per translation unit (TU) — normally one .cpp file + everything it #includes, transitively. In the pre-processing step the compiler literally substitutes…
• Cache as workspace, CPU as work — Cache as workspace, CPU as work is an interpretation discipline for dashboard signals that share a 0–100% scale but describe structurally different system behaviour.…
• Cache-coherency witness — A cache-coherency witness is a separate component that observes writes and gates reads so that a distributed cache fronting a persistent store can deliver strong read-after-write…
• Cache-control-aware grace period — A cache-control-aware grace period is the interval a publisher of rotating cryptographic key material waits between publishing a new public key and activating it as the signing…
• Cache for Availability — Cache for availability is the framing that treats a cache not primarily as a latency or bandwidth optimisation, but as a way to decouple your uptime from a dependency's uptime.…
• Cache granularity — Cache granularity is the size of the unit whose inputs form a cache key. At code level it's the function wrapped in a memoisation decorator.…
• Cache hit rate — The cache hit rate is the fraction of data requests served from the cache without needing to go to the slower backing store:
• Cache hit ratio as memory-pressure signal — The Postgres buffer-cache hit ratio — db.Cache.blkshit / (db.Cache.blkshit + db.IO.blkread) — is interpreted not as a tuning knob to maximise but as a binary memory-pressure signal…
• Cache locality — Cache locality is the property that requests for the same key consistently arrive at the same node, so a node-local cache keyed by that key accumulates hits across those requests…
• Cache stampede — A cache stampede (also called "cache-miss stampede" or "dogpile") is the specific thundering herd shape that occurs at a shared cache boundary when a popular cached entry goes…
• Cache TTL staleness dilemma — The cache TTL staleness dilemma is the forced either/or that TTL-based caching creates for rapidly-changing data: every operator has to pick one of two uncomfortable trade-offs,…
• Cache-variant explosion — Cache-variant explosion is the cartesian-product storage-pressure + hit-rate-drop phenomenon that occurs when a cache key is extended with new dimensions.…
• Callback-slice interpreter — A callback-slice interpreter is a bytecode-VM design that doesn't emit bytecode at all. The compiler emits a slice of function pointers (or closures) — one per instruction…
• Camera metadata normalization — Camera metadata normalization is the ingest-time step of taking the per-format, per-manufacturer metadata embedded in original camera files (OCF) — reel name, timecode,…
• Cancel-before-cutover — Cancel-before-cutover is the property that a staged- but-not-yet-sealed multi-migration deployment can be cancelled at any moment up to the cut-over freeze point with zero visible…
• Candidate survival rate — Candidate survival rate is the share of candidates retrieved by a candidate generator (CG) that survive the downstream ranking funnel and reach actual impression — i.e.…
• Canonical tag — The canonical tag is an HTML element — — defined in RFC 6596 (The Canonical Link Relation,…
• Canonical URL unreliability — Canonical URL unreliability is the observation that metadata-declared canonicality — specifically, the tag that a web page uses to announce its own canonical…
• XML canonicalisation — XML canonicalisation (canonical XML; C14N) is the process of converting an XML document or fragment to a byte-exact normal form so that logically-equivalent XML documents…
• Capability-based sandbox — A capability-based sandbox is a code-execution environment that starts with no ambient authority — no network access, no filesystem access, no secrets,…
• Capability manifest — A capability manifest is an upfront, declarative, install-time-inspectable statement of the capabilities a piece of third-party code (plugin, agent tool, LLM-generated extension,…
• Capacity efficiency — Capacity efficiency is the engineering discipline of reducing compute, memory, power, and capacity demand per unit of product value at hyperscale.…
• Capacity vs rate-limit quota — Two structurally different quota kinds, commonly conflated:
• Capital-localized Internet restoration — A capital-localized Internet restoration is the recurring shape in which the first phase of recovery from a government-directed shutdown is disproportionately concentrated…
• Capped VBR (QVBR) — Capped VBR — the AWS Elemental spelling is QVBR (Quality-Defined Variable Bitrate) — is a VBR rate-control variant that adds a hard instantaneous-bitrate ceiling to pure…
• Cardinality estimation — Cardinality estimation is the query-optimizer task of predicting the number of rows a sub-plan will produce before executing it.…
• Carrier thread — A carrier thread is the OS-backed worker thread in the JVM's virtual-thread fork-join pool on which a virtual thread mounts to execute. The term is defined in JEP 444.
• Cascaded LLM generation — Cascaded LLM generation is the architectural pattern of decomposing a single generative task into multiple sequential LLM phases, each with a narrower input, a targeted prompt,…
• Cascades (LLM inference) — Cascades (in the LLM-inference-serving sense) is a latency / cost-optimization technique that routes each request through a small,…
• Cascading failure — A failure that grows over time via a positive feedback loop: one node's overload spreads load to remaining nodes, increasing their probability of failure, which shifts more load,…
• Cassandra CDC commit log — The Cassandra CDC (Change Data Capture) commit log is the per-node on-disk log of mutations to CDC-enabled tables. External CDC consumers tail these files to extract row-level…
• Catalog bloat (multi-tenant table-per-tenant) — Catalog bloat is a Postgres failure mode where each backend process caches its own copy of the system catalog (pgclass, pgattribute, pgindex,…
• Catalog-managed commits — Catalog-managed commits are the architectural pattern in which every write to an open-format table (Delta, Iceberg, Hudi) is mediated by a catalog service rather than written…
• Catastrophic forgetting — Catastrophic forgetting (or "catastrophic interference") is the failure mode in which a neural network, while being trained on a new task or distribution,…
• Category-attribute relevance mapping — A category-attribute relevance mapping is an internal business-rule layer that encodes which product attributes should be shown (or extracted) for which product categories.…
• CDC bad-data propagation — CDC bad-data propagation is the structural hazard of any change-data-capture pipeline: because the target-table state at time T+1 is computed from the target-table state at time T…
• CDN as critical Internet infrastructure — Edge CDN / proxy operators like Cloudflare, Fastly, CloudFront, and Akamai sit in front of a large fraction of the world's web traffic.…
• Cell-based architecture — Cell-based architecture is the design pattern of partitioning a service into multiple independent deployable units ("cells") — each with its own compute, storage,…
• Central-first sharded architecture — Central-first sharded architecture pairs a global coordinator with regional shards: the global presents a single interface to clients (engineers, agents, APIs),…
• Centralized AI governance — Centralized AI governance is the pattern-concept of routing all of an organisation's AI traffic (LLM calls + tool calls + MCP traffic + coding-agent activity) through one policy…
• Centralized network inspection — Centralized network inspection is the architectural shape where a single managed firewall / IDS/IPS point sits between all traffic-producing networks (VPCs, on-prem data centres,…
• Certification as metadata not enforcement — Certification as metadata not enforcement is the design property that review status of a registered artifact — security approval, API-contract validation, PII-handling check,…
• cGroup ID — The cgroup ID is the 64-bit kernel-internal identifier for a cGroup node in the cgroup v2 hierarchy, stable for the lifetime of the cgroup.…
• Chained-HMAC construction — The chained-HMAC construction is the cryptographic core of Macaroons: a token's integrity tag is an iterative HMAC computation where each caveat added to the token rolls the tag…
• Change Data Capture (CDC) — Change Data Capture (CDC) is the practice of materialising a table as an ongoing stream of insert / update / delete deltas rather than (or alongside) an authoritative "current…
• Change management — Change management is the organizational process by which a production environment's intended mutations are proposed, reviewed, approved, documented, scheduled, executed,…
• Change-triggered incident rate — Change-triggered incident rate is the fraction of an organisation's customer-facing incidents that are caused by the organisation's own changes — code deploys, config rollouts,…
• Changelog as secondary index — A CDC changelog's load-bearing job description is: be a secondary index on the base table, keyed on (modified-timestamp, item-id),…
• Chaos engineering — Chaos engineering is the discipline of continuously inducing controlled failures in a production system to verify that its fault-tolerance design actually works.…
• Character set — A character set defines the set of characters that can legally be stored in a string column, plus the byte-length encoding for each character.…
• Cheapest-capable model routing — Cheapest-capable model routing is the per-task discipline of classifying every AI-agent task and routing it to the cheapest model that can actually do the task,…
• Check-then-act race — A check-then-act race is the concurrency hazard that arises when a single logical operation is implemented as two or more non-atomic steps — check whether some condition holds,…
• Checkpoint Intermediate DataFrame for Debugging — Checkpointing intermediate DataFrames is the technique of materialising a distributed Spark DataFrame to durable storage (typically S3 / HDFS / scratch paths) so it can be…
• Checkpointed automation script — A checkpointed automation script is an upgrade / migration driver that runs as a sequence of discrete steps, each of which can either:
• Checksum-validated data migration — Checksum-validated data migration is the discipline of computing a deterministic checksum of source data before a migration step (split, copy, archive, re-shard, transform),…
• Cherry-picked instrumentation payload — Cherry-picked instrumentation payload is the discipline of selecting only the observability packages that have to run in the critical page-load path and deferring everything else…
• Chrome Trace Event Format — Chrome Trace Event Format (often just "Trace Event Format") is Chromium's JSON-based trace-file format, originally documented here.…
• CI reliability — CI reliability is the fraction of continuous-integration pipeline runs that succeed for reasons attributable to the code under test.…
• Ciphersuite bloat — Ciphersuite bloat is the failure mode where a cryptographic protocol accumulates too many ciphersuites — driven typically by vendor-specific extensions,…
• Circular dependency (deployment context) — A circular dependency in the deployment context is the failure mode where the act of deploying a fix for a service depends — directly or indirectly…
• Clarity over efficiency in protocol design — Design principle stating that in shared schema / protocol design, the primary objective is to leave no ambiguity for implementers — cross-team, cross-language,…
• Clean-room recovery account — A clean-room recovery account is a separate AWS account (or analogous cloud-provider isolation boundary) used as the destination for backup / replication,…
• Clear-text name resolution — The bidirectional translation between a messaging platform's opaque identifiers (U07ABCDEF, #C04DEFGHI) and human-readable names (@alice, #engineering),…
• CLI convention enforcement — CLI convention enforcement means baking naming and flag conventions into the schema layer of a CLI's code-gen pipeline so that inconsistency becomes a compile-time impossibility,…
• Click-duration reweighting — Click-duration reweighting is a training-data preprocessing technique that attenuates noisy binary click labels by scaling their training weight with a monotonic function of click…
• ClickHouse append-only tamper resistance — ClickHouse's MergeTree engine is append-friendly by design — it does not support row-level UPDATE or DELETE in the normal path.…
• ClickHouse data part — A data part (or simply part) is the canonical physical storage unit in ClickHouse's MergeTree family of table engines. A part is an immutable, self-contained,…
• ClickHouse detached partition archival — ALTER TABLE DETACH PARTITION '' is the ClickHouse mechanism for physically removing a whole partition from the active queryable dataset while keeping the underlying data…
• ClickHouse distributed query — In ClickHouse, cross-shard queries route through distributed tables (backed by the Distributed table engine in a conventionally-named default database).…
• ClickHouse index granule skipping — ClickHouse MergeTree's primary index is a sparse index of ORDER BY-key minimum values stored once per granule* (a block of rows, default 8,192 rows).…
• ClickHouse MergeTree partition by time — The canonical idiom for ingesting time-series data into a ClickHouse MergeTree table: declare PARTITION BY toYYYYMM(timestamp) (monthly) or PARTITION BY toDate(timestamp) (daily)…
• ClickHouse trace_log — system.tracelog is a built-in ClickHouse system table that records stack-trace samples from the running ClickHouse server.…
• ClickHouse TTL policy — A ClickHouse TTL policy is a table- or column-level clause that declaratively specifies when rows or columns should be automatically deleted or moved to a different storage tier…
• Client-middleware interception — Client-middleware interception is the use of a producer-side middleware chain — code that runs in the caller's process at the moment an operation is initiated,…
• Client release needed only for new primitives — In a server-driven UI system that adopts UI primitives as platform building blocks and server-owned schema versioning, the release boundary collapses to a single explicit…
• Client-server decoupling — Client-server decoupling is the architectural pattern of splitting a monolithic distributed-compute driver / coordinator into a client (user application code,…
• Client-server model — The client-server model is the foundational deployment pattern of the Internet: a client makes a request to a server, which responds with a resource.…
• Client-side black-box probe — A client-side black-box probe is a synthetic-monitoring primitive that issues real, protocol-shaped requests from outside the service-under-test and reports pass/fail + timing…
• Client-side compression — Client-side compression is the choice to compress payloads at the client before sending them over the network (and likewise decompress server responses on the client),…
• Client-side load balancing — Client-side load balancing means the caller chooses which backend instance to send a request to, rather than delegating that decision to a proxy (L7 sidecar,…
• Client-side observability — Client-side observability is the practice of emitting tracing and metrics data from the browser — not only from the server path…
• Client-side performance instrumentation — Client-side performance instrumentation is the practice of emitting performance timestamps and metrics from inside the user's device — the mobile app, browser,…
• Client-side query compression — Client-side query compression is the compression of a database query's request payload (or batch of payloads) by the client before it crosses the network,…
• Client-side removal of noindex is ineffective — The operational rule: if a page's initial HTML response — the bytes the server emits before JavaScript runs — contains (or an equivalent…
• Client spec version — A client spec version is a bundled-with-the-client manifest that lists, as data, the versions of every component interface the client can handle.…
• Cloaking — Cloaking is the practice of serving different content to users and search engines for the same URL, typically by branching on the User-Agent header (or source IP / ASN /…
• closeWait socket state — CLOSEWAIT is a TCP endpoint state: the remote peer has sent a FIN (it closed its side), but the local application has not yet called close() (or shutdown(SHUTWR)) on the socket.…
• Cloud IDE — A cloud IDE is a development environment delivered through the browser whose actual execution — file system, shell, compiler, runtime,…
• Cloudflare zone analytics — Cloudflare zone analytics is the per-zone (per-customer- domain) record of which URLs received traffic over a recent window,…
• CloudFormation ChangeSet — A CloudFormation ChangeSet is AWS CloudFormation's dry-run API for a stack update. Given an updated template (or parameters),…
• CloudFormation Verbosity Problem — The CloudFormation verbosity problem is the operational pain that AWS CloudFormation templates — even when expressive enough to describe any AWS resource set…
• Cluster epoch — A cluster epoch is a monotonically increasing, cluster-global counter that is embedded into the durable identifier of an object at the moment it is created,…
• Cluster health check — A cluster health check is a liveness / readiness probe that a router (load balancer, gateway, proxy) issues against a backend cluster to decide whether to send it traffic.…
• Cluster-level tenant isolation — Cluster-level tenant isolation names the isolation grain where the compute cluster itself is the tenant boundary — each tenant gets a dedicated scheduler-managed cluster (an ECS…
• Clustered index — A clustered index is a database index in which the table itself is the index: rows live inside the index structure keyed on the primary key, in primary-key order.…
• Customer-Managed Keys (CMK) — Customer-Managed Keys (CMK) is the encryption-key ownership model where the customer (the data-owning organisation) controls the root-of-trust key in their own cloud KMS…
• CMS entry-type modular composition — Entry-type modular composition is a CMS content-model pattern where a top-level entry (a "page") is a thin container holding a list of references to module entries,…
• CNAME chain — A CNAME chain is a sequence of DNS alias records traversed during name resolution:
• Co-clustered join — A co-clustered join is a join optimisation that eliminates data shuffle when both join sides are clustered on the join key.…
• CO-STAR prompt framework — CO-STAR is a six-part prompt-authoring framework popularised in 2024:
• Code as context for data agents — The architectural insight that for AI data agents (LLM agents that turn natural-language questions into SQL), the transformation code that produces a table carries semantic…
• Code indexing — Code indexing is the job of collecting structured facts from source code so that developer tools can answer questions efficiently…
• Code-vs-label mismatch as catalog audit — Code-vs-label mismatch as catalog audit is the technique of treating disagreement between a learned-from-features Semantic ID cluster and a manually-assigned taxonomy label…
• Coding-agent sprawl — Coding-agent sprawl is the named problem class of an engineering organisation simultaneously running multiple parallel coding-agent tools (Cursor, Claude Code, Codex CLI,…
• Cold-start connection cost — Cold-start connection cost is the latency paid before the first query byte flows on a newly-opened database (or service) connection…
• Cold Start — Cold start names two distinct phenomena that share a name because both describe "the first time around is slow / hard":
• Cold-storage minimum-duration tax — The cold-storage minimum-duration tax is the structural failure mode of cold-tier object-storage classes (e.g. S3 Glacier family, Azure Archive,…
• Collation — A collation is a set of rules for comparing and sorting strings. Every string character set in MySQL has at least one collation,…
• Collective-communication topology awareness — Collective-communication topology awareness is the practice of choosing collective-operation algorithms (all-reduce, all-gather, reduce-scatter,…
• Column-level access control — Column-level access control enforces data governance at the granularity of individual columns in a dataset, rather than at the table or database level.…
• Column-order conflict — A column-order conflict is a schema-merge conflict class where two concurrent deploy branches each add a distinct column to the same table,…
• Columnar storage format — A columnar storage format lays out a tabular dataset on disk by column rather than by row: all values of column A are contiguous, then all values of column B, etc.…
• Commit-graph bisection — Commit-graph bisection is a VCS-side binary-search technique that answers queries like "when did this file's content change?" (log) or "who last touched line L?" (blame) in O(log…
• Commit Sequence Number (CSN) — A Commit Sequence Number (CSN) is a monotonically-increasing integer stamped onto a transaction at commit time, used as the canonical position of that commit in the database's…
• Commit signing — Commit signing is the cryptographic attachment of a signature to a Git commit (or tag) that lets verifiers later confirm the commit's contents and/or origin were not tampered…
• Commit stack — A commit stack is a sequence of small, incremental, individually- reviewable commits built on top of each other as a single feature's worth of work…
• Common-deadline scheduling — Common-deadline scheduling is the problem-class restriction where "all jobs share a common deadline (e.g., all data processing must finish by the nightly batch run)" (Source:…
• Compass, not encyclopedia — "Compass, not encyclopedia" is the design principle Meta names for per-module AI-agent context files: each file is 25-35 lines / ~1,000 tokens…
• Competitive ratio — Competitive ratio is the canonical quality metric for an online algorithm: "the worst case comparison between the throughput of our online algorithm and the throughput…
• Compiler-generated race condition — A compiler-generated race condition is a data race whose root cause is not in the user's source code but in the sequence of machine opcodes the compiler or assembler emitted…
• Component-scoped field access — Component-scoped field access is a GraphQL schema discipline that restricts a field to a named allowlist of UI components (or equivalently, named queries/operations).…
• Component version migrate function — A component version migrate function is an explicit, per-component function — typically auto-generated as a stub but implemented by developers…
• Composable ML modules — Composable ML modules are self-contained, versioned executable units that encapsulate a single ML concern (data processing, model training, evaluation,…
• Composable tag axes — Composable tag axes is the design property of a query- tagging rule engine that allows multiple orthogonal tag dimensions to coexist on the same query,…
• Composite fingerprint signal — A composite fingerprint signal is a detection construct that blocks or throttles a request only when two independent classes of evidence agree…
• Composite-index column order — The order in which columns appear in a composite index definition determines which queries can use the index and how efficiently…
• Composite index — A composite index (a.k.a. multi-column index, compound index) is a single database index whose key is a tuple of two or more columns in a specified order.…
• Composite-key upsert — A composite-key upsert is a write operation that uses a multi-field-derived unique key (e.g. (assetid, timebucket)) as the stable document identifier such that a subsequent write…
• Compositional specification — A compositional (or modular) specification describes a system as a set of smaller specs connected by explicit interface boundaries,…
• Compression codec trade-off — Compression codec trade-off is the choice a streaming producer makes between space / bandwidth savings (high compression ratio = fewer bytes over the wire, less disk,…
• Compression + compaction CPU cost — Compression + compaction CPU cost names the broker-side CPU tax that arises when a topic is both compressed (record data is encoded with ZSTD / LZ4 / gzip / snappy) and compacted…
• Compression Plan — A compression Plan is the per-use-case configuration object that an OpenZL trainer produces from a shape description (via SDDL or a registered parser function) and sample data.…
• Compression side-channel attack — A compression side-channel attack exploits the fact that the size of a compressed payload depends on the content being compressed — specifically,…
• Compromise-boundary recovery point selection — Compromise-boundary recovery point selection is the recovery-time algorithm for choosing which backup to restore when a cyber event has been confirmed.…
• Computational backtest — A computational backtest is an offline evaluation of a decision-making system (trading algorithm, replenishment engine, recommender) by replaying historical data through the system…
• Compute shader — A compute shader is a GPU program that performs general compute work independent of the fixed-function rendering pipeline.…
• Compute–storage separation — Compute–storage separation is the architectural property where a system's persistence layer and its query/compute layer are decoupled and scale independently.…
• Computed Pattern — Computed Pattern — MongoDB's named schema-design pattern where values that would otherwise be computed at read time (sums, counts, averages,…
• Concentration risk on single-cloud LLM serving — Concentration risk on single-cloud LLM serving is the structural reliability failure mode of running production LLM-powered features against a single cloud provider's serving…
• Concurrent rendering (React 18) — Concurrent rendering is the React 18 feature set that decouples deciding what to render from committing it to the DOM, allowing the renderer to pause, resume, prioritise,…
• Condition-tag IAM — Condition-tag IAM is the AWS IAM authorization pattern where an IAM policy's effect on a request is gated on AWS resource tags via the Condition element,…
• Confidence-tiered routing — Confidence-tiered routing is the discipline of slicing AI decision confidence into three (or more) bands, each with a distinct downstream treatment,…
• Confidential computing — Confidential computing is the posture of protecting data in use — i.e. plaintext that is being actively computed on — via hardware-enforced isolation primitives (TEEs).…
• Config-as-code pipeline — A config-as-code pipeline is a data / infrastructure pipeline whose behaviour is primarily driven by code-versioned configuration files (Python / YAML / Hack / Starlark / …)…
• ConfigMap hash rollout — ConfigMap hash rollout is the mechanism by which Kustomize forces a rolling restart of pods that mount a ConfigMap by appending a content-hash suffix to the ConfigMap name,…
• Configuration bloat — Configuration bloat is a multi-tenant SaaS failure mode where a tenant's accumulated configuration entities (custom fields, work types, screens, schemes, workflows,…
• Conflict-free Replicated Data Type (CRDT) — A Conflict-free Replicated Data Type (CRDT) is a data structure replicated across multiple machines such that: (1) any replica may be updated independently and concurrently without…
• Conflict-resolution memoization — Conflict-resolution memoization is the technique of recording the human-resolved outcome of a merge or cherry-pick conflict and automatically re-applying that resolution when…
• Conformance checking — Conformance checking is the problem of mechanically verifying that a production implementation behaves as a formal specification (TLA+, Alloy, Event-B,…
• Congestion window — The congestion window (often cwnd) is the transport- protocol-level cap on how many bytes the sender may have in flight but unacknowledged at any moment.…
• Connection failure rate — Connection failure rate is a metric measuring how often a reverse proxy or CDN edge fails to establish a successful connection to an origin server when attempting to retrieve…
• Connection multiplexing over HTTP — Connection multiplexing over HTTP is the subset of connection multiplexing that uses HTTP/2 or HTTP/3 streams as the multiplexing substrate…
• Connection multiplexing — Connection multiplexing is the architectural pattern of decoupling the number of upstream connections a backend holds open from the number of client-side connections by interposing…
• Connection pool exhaustion — A connection pool is a fixed-size set of pre-established database connections shared across application workers. An application worker takes a connection from the pool…
• Connection time — Connection time is the wall-clock time for an end-user device to complete the handshake with the endpoint it's trying to reach…
• Connection timeout — A connection timeout is the maximum time a client waits while attempting to establish a network connection with a server.…
• Conservative extension — In logic + formal methods, a theory T' is a conservative extension of a theory T when:
• Consistency threshold per endpoint — Consistency threshold per endpoint is the parallel-run readiness criterion: each endpoint (identified by operationid in Zalando's variant) gets its own target match-rate percentage…
• Consistent caching for horizontal-scale read APIs — A pattern for taking a singleton leader-elected source-of-truth service — where the data fits in memory and latency demand is low…
• Consistent hashing — Consistent hashing maps keys to buckets (shards, servers, cohorts) in a way that minimises re-mapping when the bucket set changes.…
• Consistent identifier naming — A data architecture principle requiring that the structure of primary identifiers is directly dependent on the modeling choice.…
• Consistent Lookup Vindex — A Consistent Lookup Vindex is a Vitess Secondary Vindex variant that keeps user-facing query results consistent with the authoritative data table without running 2PC (Two-Phase…
• Consistent non-locking snapshot — A consistent non-locking snapshot is a point-in-time view of a database (or a table) from which large amounts of data can be read without blocking concurrent writes and without…
• Consistent read — A consistent read is a read that observes every write committed before it — no stale state, no lost acknowledged updates.…
• Consolidated billing with volume discount — Consolidated billing is the AWS Organizations primitive that rolls charges from every member account up to a single payer account and applies AWS's tiered- pricing volume discounts…
• Constant bitrate (CBR) — Constant bitrate (CBR) is a video-encoder rate-control mode in which the encoder targets a fixed bytes/sec output, regardless of scene complexity.…
• Constant work pattern — Constant work is Colm MacCárthaigh's named AWS reliability pattern: build systems that do the same amount of work regardless of load.…
• Constrained decoding (structured output) — Constrained decoding is the family of LLM inference techniques that restrict the set of tokens the model can emit at each generation step,…
• Constrained-resource benchmark — A constrained-resource benchmark is a benchmark run with an explicit resource cap (CPU cores, memory, IOPS, network bandwidth) chosen to represent a specific production tier,…
• Consul streaming vs. long-polling — A concrete case study in how a concurrency-primitive-design choice made during performance optimization regressed catastrophically under extreme load.…
• Consumer fetch tuning — Consumer fetch tuning is the Kafka / Redpanda client-side practice of explicitly configuring the four fetch parameters that govern how much data a consumer pulls per broker fetch…
• Consumer group — A consumer group in Apache Kafka is a set of consumers that cooperatively divide the partitions of a topic among themselves.…
• Contagious deadlock — A contagious deadlock is a concurrency bug that propagates via the shared data plane: when the deadlock-triggering input reaches a new node, that node also deadlocks.…
• Container-entrypoint compatibility layer — A container-entrypoint compatibility layer is code that runs inside a Docker container's entrypoint script (before the user's command),…
• Container ephemerality — Container ephemerality is the operational-semantics property that a container instance's filesystem state is not durable past the container's lifetime — deletion, replacement,…
• Container escape — A container escape is when code inside a container breaks through the isolation boundary and obtains execution or data access on the host (or on sibling containers on the same…
• Container-extracted SBOM — A container-extracted SBOM is a Software Bill of Materials produced by scanning the built container image rather than the source tree or package lock-files.…
• Content-addressed caching (of build/test actions) — Content-addressed caching stores action outputs keyed by a hash of the action's inputs (source contents, tool versions, environment, dependencies, flags).…
• Content-addressed ID — A content-addressed ID is an identifier derived deterministically from the content of a record via a cryptographic hash, so that two records with identical content necessarily…
• Content-grounded answer — A content-grounded answer is an LLM-product discipline in which every answer the system returns must be supported by evidence retrieved from a vetted content source…
• Content ID fingerprint — A content ID fingerprint is a function over a rendered page (or more generally, any content artefact) that returns the same identifier for two artefacts iff their observable…
• Content management as platform — Content management as platform is the architectural stance that editorial content management is a platform capability owned by a platform-engineering team…
• Content Signals — Content Signals is a proposed extension to robots.txt that lets a site declare what AI systems may do with its content, separately for three orthogonal dimensions:
• Context as importance signal — Context as importance signal is the observation that anything loaded into an LLM's context window is read by the model not just as capability available but also…
• Context-aware retrieval — Context-aware retrieval is the RAG refinement where the retrieval query is enriched with case-specific structural metadata (tenant, jurisdiction, document type, risk level,…
• Context cancellation ignored in retry loop — The context-cancellation-ignored-in-retry-loop bug is a concurrency defect in which a retry loop — typically inside an operator's reconcile path,…
• Context compaction — Context compaction is the practice of trimming or summarising older content in an LLM agent's context window to keep a long-running reasoning loop within token limits without…
• Context-encoded LLM prompt — A context-encoded LLM prompt is a prompt whose template includes structured, environment-specific facts (data sources, custom interpreters, configuration patterns,…
• Context engineering — Context engineering is the discipline of allocating a fixed token budget across the components that compete for the LLM's context window — system prompts, tool descriptions,…
• Context exhaustion — The failure mode where an LLM agent's context window fills up and the model begins cannibalising its own memory — forgetting bugs it spent hours tracking down,…
• Context file freshness — Context file freshness is the discipline of keeping AI-agent context files in sync with the underlying code they describe so that references (file paths, function names,…
• Context layer in two-tower — A context layer is a tower-internal architectural component inside a two-tower retrieval / ranking model that consumes real-time, request-time signals (the current page,…
• Context-propagated SQL tags — Context-propagated SQL tags is the idiom for attaching per-request metadata to SQL queries in languages without an implicit request-local storage convention — most canonically Go,…
• Context propagation (OpenTelemetry) — Context propagation is the OpenTelemetry mechanism that carries trace/span identifiers across service boundaries — so a request that hops through many services produces a single…
• Context rot — Context rot is the empirically observed degradation of LLM agent accuracy as the context window fills up over a long-running investigation…
• Context switch — A context switch is the OS kernel's act of saving the current process (or thread)'s execution state and restoring another one's, so the CPU can resume a different unit of work.…
• Context window as token budget — The context window supplied to an LLM call is a fixed token budget. Every input the program keeps in that window — user messages, assistant replies,…
• 'continue' prompt for truncated output — The "continue"-prompt primitive is the idiom of sending the literal string "continue" on the same conversation turn when an LLM response is cut off at the output-token limit,…
• Continued pretraining — Continued pretraining (also called "continual pretraining" or "domain-adaptive pretraining") is the technique of taking an already-pretrained foundation model and running…
• Continuous-computation convergence (batch + streaming) — Continuous-computation convergence is the observation that the decades-long distinction between batch and streaming data processing collapses once:
• Continuous profiling — Continuous profiling is the telemetry signal class that tells you which function, on which line, is burning CPU or memory in production — continuously,…
• Continuous reprediction — Continuous reprediction is the online policy of re-running a predictor as new observations arrive, replacing a single commit-at-creation prediction with a stream of updated…
• Continuous schema deployment — Continuous schema deployment is the contract where schema migrations are treated as a routine, high-frequency operation — shipped multiple times per day, per engineer,…
• Contract-first design — Contract-first design = defining the API contract (endpoints, request/response shapes, error codes, auth) before any implementation,…
• Contract testing — Contract testing verifies an API contract from both sides: the consumer writes tests against its own expectations of the provider's schema,…
• contract.yaml as bus surface — contract.yaml as bus surface is the OmniNode discipline of making each node ship a YAML manifest that owns the node's event-bus surface…
• Contrastive regularization with catalog structure — Contrastive regularization with catalog structure is the technique of adding a contrastive-loss term to a representation-learning objective (e.g.…
• Control plane is the new data plane — "Control plane is the new data plane" is the architectural claim that under agentic, serverless, scale-to-zero workloads,…
• Control-plane change blast radius — Control-plane change blast radius is a specialised framing of blast radius that distinguishes between:
• Control plane / data plane separation — Architectural split between the "decide" path (control plane: validation, authorization, policy, rollout decisions, scheduling) and the "deliver" path (data plane: storage,…
• Control-plane fan-out to the Kubernetes API — Control-plane fan-out to the Kubernetes API is the scaling anti-pattern where N identical data-plane pods each independently poll or watch the API server for the same resources,…
• Control-plane impact without data-plane impact — Control-plane impact without data-plane impact is the specific operational outcome-shape in which a system's configuration / orchestration path is fully broken but the data path…
• Conversion-rate uplift — Conversion-rate uplift (CVR uplift) is the standard e-commerce / search-ranking A/B testing metric: the absolute or relative difference in conversion rate between the treatment…
• Conway's Law — Conway's Law (Melvin Conway, 1968) is the sociotechnical observation that systems tend to mirror the communication structure of the organizations that build them.…
• Cookbook artifact versioning — Cookbook artifact versioning is the Chef-ecosystem unit of configuration-management rollout: a pinned-version cookbook artifact (typically a .tar.gz of the cookbook tree + its…
• Coordinated bot network — A coordinated bot network is a fleet of browser-automation instances controlled by a single operator that share operator-controlled properties…
• Coordinated disclosure — Coordinated disclosure (historically responsible disclosure) is the industry norm by which a security vulnerability is not made public until the affected vendor has had…
• Copy Fail — CVE-2026-31431 — Copy Fail (CVE-2026-31431) is a Linux kernel local-privilege-escalation vulnerability disclosed publicly by Xint Code on 2026-04-29.…
• Copy-on-write merge (compaction) — Copy-on-write merge is the compaction strategy that collapses a stream of CDC deltas (insert / update / delete) into a new,…
• Copy-on-write storage fork — A copy-on-write storage fork is a storage-cloning mechanism that creates a second logical copy of a dataset without initially duplicating the underlying pages…
• Core Web Vitals — Core Web Vitals is Google's standardised set of user- experience metrics for web pages, measured in the browser and intended as proxies for perceived page quality (loading,…
• Correlated EBS failure within an availability zone — Correlated EBS failure within an AZ is the observation that EBS volume performance degradation is not independent across volumes — a single underlying networking, control-plane,…
• Cosine similarity — Cosine similarity is the cosine of the angle between two vectors:
• Cosmetic logout — Cosmetic logout is the vulnerability where a user-visible "log out" action clears local state (cookies, UI) but does not actually revoke the underlying token at the authority…
• Cost tracking per team — Cost tracking per team is the platform-engineering discipline of attributing every inference call (and its dollar cost) back to the consuming internal team,…
• Cost-weighted deletion deferral — Cost-weighted deletion deferral is the policy that resources with low cost-savings potential relative to the impact of a wrong deletion are excluded from automated cleanup…
• Counterfactual outcome prediction — Counterfactual outcome prediction is the use of an ML model to estimate what an outcome would have been under a treatment that wasn't actually applied.…
• Counterfactual stockout demand modeling — When an article stocks out mid-period, the only data you observe is "sold X, where X ≤ on-hand inventory". The true demand…
• Coupled vs decoupled database-schema and application deploy — A coupled deploy ships application-code changes and database-schema changes together, as a single release unit. A decoupled deploy ships the two changes in separate,…
• Coverage cell — An (area × attack-class) matrix cell used to measure and track vulnerability scanning completeness across a codebase.
• Covering index — A covering index is a secondary index that contains every column the query projects — so the query can be answered entirely from the index,…
• CPU-bound serving regime for small fast models — The CPU-bound serving regime is the inference operating point at which a single CPU process can no longer prepare the next batch fast enough to keep the GPU saturated,…
• CPU busy-loop incident — A CPU busy-loop incident is the recurring operational shape where one or more processes on production hosts peg a core at near-100% with little corresponding useful work,…
• CPU cache hierarchy (L1 / L2 / L3) — Modern CPUs do not read directly from RAM. They read from a hierarchy of small, fast on-die caches that sit between the execution units and main memory:
• CPU-only media processing — CPU-only media processing is the deliberate choice to run image / video / audio processing workloads on CPU instances in the cloud,…
• CPU starvation of network driver threads — A class of incident where a network driver's kernel thread does not get scheduled onto a CPU core for multiple seconds, during which the hardware driver concludes the device…
• CPU throttling vs noisy neighbor — The two distinct scheduler pathologies that present identically in run-queue latency, and which therefore cannot be told apart by that metric alone.
• CPU time breakdown — Linux accounts CPU time across a fixed set of named buckets in /proc/stat, surfaced on vmstat / mpstat / top / sar as columns (us / sy / id / wa / st / hi / si / ni / guest /…
• CPU utilisation ceiling on databases — Database workloads should not run hot on CPU. Zalando's empirical observation from running an RDS Postgres fleet: CPU utilisation above 40-60% on a database instance is an incident…
• CPU utilization vs saturation — Utilisation and saturation are two separate measurements of the same CPU. They can (and frequently do) diverge, and conflating them is one of the most common triage mistakes.
• CPU vs Real flame graph — A CPU flame graph and a Real (wall-clock) flame graph are two distinct profiling modes that look superficially identical (same horizontal-stacks-of-frames visualisation) but answer…
• CQRS (Command-Query Responsibility Segregation) — CQRS — Command-Query Responsibility Segregation — is the idea that the data model optimized for accepting writes (commands) doesn't have to be the same data model serving reads…
• Crash-consistent replication — Crash-consistent replication produces recovery points whose state is equivalent to what an unplanned crash + reboot of the source system would leave on disk.…
• Crawl budget impact of JS complexity — Crawl budget is Google's shorthand for the aggregate per-site capacity allocation — how many pages Googlebot will fetch and render from a given site in a given time window.…
• CRDT (conflict-free replicated data type) — A Conflict-free Replicated Data Type is a data structure whose replicas can be updated concurrently and without coordination and will still converge to the same value…
• Credential / Route / Endpoint triple — The Credential / Route / Endpoint triple is PlanetScale's canonical data model for separating the three concerns that arise when admitting a database connection at a globally…
• Credential vending — Credential vending is the architectural pattern in which a central authority (typically a catalog, registry, or governance plane) mints short-lived,…
• Credibility scoring rubric — A credibility scoring rubric is a named, fixed-grained scoring scale — typically 5 levels — used by a critic agent to annotate peer-agent findings with numeric credibility scores…
• Critical business operation — A Critical Business Operation (CBO) is a named, top-level user-facing business action (checkout, item view, order placement, sign-in,…
• Critical-path dependency minimization — Critical-path dependency minimization is the reliability discipline of reducing the count and depth of external dependencies on the synchronous request path of a load-bearing…
• Critical path (build / pipeline / distributed DAG) — In a DAG of dependent actions (build targets, pipeline steps, tasks), the critical path is the longest chain of dependent work from the graph's root to its sink.…
• Critical User Journey (CUJ) — A Critical User Journey (CUJ) is a user-workflow-centric design primitive: the full end-to-end sequence of steps a user takes to accomplish a business-critical task,…
• Cross-account backup — Cross-account backup is backup / replication written to a different AWS account from the source, typically with distinct credentials and separate access control.…
• Cross-AZ replication bandwidth cost — Cross-AZ replication bandwidth cost is the per-byte cloud- provider charge for transferring data between two availability zones within the same region…
• Cross-Cloud Architecture — Cross-cloud architecture is the (deliberate) choice to run components of one system across more than one hyperscaler (AWS, Azure, GCP, …).…
• Cross-cloud cost consolidation — Cross-cloud cost consolidation is the re-platform strategy of moving components out of a "component A on cloud X, component B on cloud Y" topology and into a single-cloud topology.…
• Cross-cluster federated query cost — When a multi-tenant storage system is split into multiple failure- domain clusters for blast-radius reasons (see active-multi-cluster-blast-radius),…
• Cross-cluster networking — Cross-cluster networking is the set of problems that arise when a workload's components end up running in different clusters — each with its own network namespace,…
• Cross-cluster offset translation map — A cross-cluster offset translation map is an external per-consumer-group data structure that records, for every partition replicated between two Kafka-API clusters,…
• Cross Cluster Replication (CCR) — Cross Cluster Replication (CCR) is Elasticsearch's primitive for replicating index data between otherwise-independent Elasticsearch clusters.…
• Cross-Encoder Reranking — Cross-encoder reranking is a two-stage retrieval pattern where an initial fast retrieval stage returns a candidate set of ~10–100 documents,…
• Cross-engine ABAC — Cross-engine ABAC is the architectural shape in which a single set of attribute-based access control policies — defined once in a catalog…
• Cross-engine SQL data-type equivalence — Cross-engine SQL data-type equivalence is the concept underlying any schema migration between two SQL engines: two engines can both declare a POINT column and still store the data…
• Cross-lingual knowledge transfer — Cross-lingual knowledge transfer is the LLM phenomenon of applying knowledge learned during pre-training in one language to question-answering or task performance in a different…
• Cross-model portability — Cross-model portability is the property that switching the underlying model (or model provider) in a production system requires no caller-side code change…
• Cross-partition authentication — The design problem of authenticating a workload across two or more AWS partitions, given that IAM credentials are bound to a single partition and don't carry across the partition…
• Cross-payment-method signal propagation — Cross-payment-method signal propagation is the architectural mechanism where a fraud signal detected on one payment method (e.g.…
• Cross-platform client library — A cross-platform client library is a single shared implementation of a capability (media processing, TLS, crypto, state sync,…
• Cross-platform preference parity — Cross-platform preference parity is the discipline of maintaining a single shared preference model across all client platforms (desktop, iOS, Android,…
• Cross-platform UI subset tradeoff — Any cross-platform UI abstraction — regardless of which substrate it picks (HTML, React Native primitives, Flutter widgets, SwiftUI,…
• Cross-Region backup — Cross-Region backup is backup / replication written to a different AWS Region from the source. The target Region is a fault-isolation boundary…
• Cross-region bandwidth cost — Cross-region bandwidth cost is the per-byte cloud-provider charge for transferring data between two geographic regions of the same provider (e.g., AWS us-east-1 ↔ eu-west-1,…
• Cross-regional attribution via CIDR trie — When workloads span multiple cloud regions, cross-regional IP attribution must resolve a remote IP that may belong to a workload in a different region. Two shapes are possible:
• Cross-shard query — A cross-shard query is any query whose execution requires data from more than one physical shard in a horizontally-sharded database.…
• Cross-signed certificate trust (double-signed certificates) — A PKI design where two (or more) isolated root certificate authorities cross-sign each other's certificates, producing a bidirectional chain of trust so relying parties in either…
• Cross-surface content unification — Cross-surface content unification is the property that a single piece of domain data — rendered in multiple product surfaces (landing page, product detail page, email campaign,…
• Cross-VPC private connectivity — Cross-VPC private connectivity is the problem of giving an external system — another VPC, another account, another cloud,…
• Cryptographic inventory (crypto-inventorying) — Cryptographic inventory is the organisation-wide mapping of where cryptographic primitives are used — which algorithms, which keys, which call-sites, which services,…
• Cryptographic monitoring — Cryptographic monitoring is the practice of logging every cryptographic operation performed across a fleet — the key name, the algorithm,…
• Cryptographic Shredding — Cryptographic shredding is the technique of rendering data inaccessible by destroying (or denying access to) the encryption keys that decrypt it,…
• Cryptographically-relevant quantum computer (CRQC) — A cryptographically-relevant quantum computer is a quantum computer scaled to the point where it can execute Shor's algorithm on key sizes in actual production use…
• CSS subset cross-platform UI — CSS subset cross-platform UI is the approach of writing application UI in an HTML + CSS subset that a library maps to each target platform's native primitives at build/runtime.…
• CTR prediction — CTR prediction (click-through rate prediction) is the machine-learning task of estimating the probability that a user will engage (click, like, watch,…
• CUBIC congestion control — CUBIC is a TCP congestion-control algorithm introduced in 2008 and standardised in RFC 8312 (2018). It has been the default congestion controller in the Linux kernel since Linux…
• CUBIC epoch — The epoch (epochstart in the Linux-kernel TCP CUBIC implementation; congestionrecoverystarttime in quiche) is CUBIC's reference timestamp for its growth curve.…
• CUDA stream — A CUDA stream is an in-order queue of GPU operations (kernels, memory copies, events). Operations enqueued on the same CUDA stream execute strictly in the order they were issued.…
• CUDA throughput budget — CUDA throughput budget is the GPU-throughput cost profile of a serving workload — the queries-per-second per GPU (or equivalently per-request GPU time) that a given model consumes…
• Curse of the last reducer — The curse of the last reducer is the canonical name for the failure mode in batch (MapReduce / Spark / Flink batch) jobs where the slowest partition / reducer dominates wall-clock…
• Cursor pagination — Cursor pagination (also known as keyset pagination or seek-method pagination) is a pagination mechanism that uses the value of a sort-key column from the last seen row…
• Custom histogram buckets — Custom histogram buckets are explicit bucket boundaries passed to a metrics histogram instrument, overriding the library default,…
• Customer-driven metrics — Customer-driven metrics are workload measurements that are driven by the customer's behaviour, independent of the infrastructure's response — queries per second,…
• Customer-facing SLA — A customer-facing SLA is a quantitative promise stated in metrics the customer directly perceives — not internal system metrics.…
• Customer impact hours metric — A customer impact hours metric is a reliability program's top-line metric defined as summed hours of customer-visible impact from high-severity and filtered medium-severity…
• Customer-intent-driven experience selection — The principle that a request-scoped Experience (see experience-zalando-if) is selected from explicit customer intent — what the customer asked for,…
• Customer-managed IAM footprint minimization — Customer-managed IAM footprint minimization is the architectural principle where a vendor's managed-service offering operates with the smallest possible IAM privilege set inside…
• Customer-provider relationship (BGP) — A customer-provider relationship is one of the two primary pairwise business relationships between ASes: the customer pays the provider for Internet transit.…
• Customer Zero — Customer Zero is the discipline of running a platform team's own first-party product on the platform's own substrate before external customers do,…
• Cut-over freeze point — The cut-over freeze point is the single brief moment in an online schema change (or a data-motion migration) at which the source table is made write-locked on the server so…
• Cyber frontier model — A cyber frontier model is a frontier-tier LLM specialised for offensive security work: vulnerability discovery, exploit reasoning, and proof-of-exploitability generation.…
• Cyber resilience — Cyber resilience is the recovery leg of a three-leg security posture — prevention keeps threat actors out, detection finds them quickly,…
• Cyclic foreign key prohibition — Cyclic foreign key prohibition is the restriction on Vitess's foreign-key constraint support (as of 2023-12-05) that forbids FK cycles across different tables while continuing…
• DAG id rewriting — DAG id rewriting is mutating an orchestrator workflow's identifier at load/init time so that multiple copies of the same source workflow can coexist on a single orchestrator server…
• DAG vs cyclic workflow — Most mainstream workflow orchestrators constrain their workflows to Directed Acyclic Graphs (DAGs) — a dependency graph with no loops.…
• Data agent unique challenges — A data agent answers questions over enterprise data — structured tables, dashboards, notebooks, plus unstructured workspace files, documents, and chat logs.…
• Data annotation (IFC labels) — In information flow control systems like Meta's Policy Zones, a data annotation is a metadata label attached to data assets (e.g.…
• Data branching — Data Branching® (PlanetScale trademark) is the variant of database branching where the branch is created with a copy of the production data, not just the production schema.…
• Data-centric AI governance — Data-centric AI governance is the principle — canonicalised in Databricks' 2026-05-20 four-pillars post — that an agent's behaviour is almost entirely determined by the data it has…
• Data classification tagging — Data classification tagging is attaching machine-readable sensitivity metadata to each unit of data — typically at the column / field level — so that automated policy enforcement,…
• Data container — A data container is a named repository of instance data that conforms to a specific schema language / type system and operates within a specific system's structural and operational…
• Data contract — A data contract is an explicit, versioned agreement between a data producer and one or more data consumers about the schema, semantics, quality guarantees,…
• Data debt migration — Data debt is the accumulation of legacy tables, dashboards, and pipelines that no longer meet current modeling standards but remain in use because of downstream dependencies.…
• Data-driven flushing — Data-driven flushing is the sink-connector design where output to the downstream store happens only when there is actual data to flush — not on a fixed-interval timer.…
• Data environment — A data environment is a named set of databases / tables / views (all sharing a naming suffix) that a pipeline environment reads from and writes to.…
• Data file analysis — Data file analysis is the agent primitive of scanning every file in a working directory — structured and unstructured — and emitting a rich textual description of each file's…
• Data flow violation — In IFC systems like Meta's Policy Zones, a data flow violation is the runtime event raised when annotated data would flow into a destination whose annotation (or lack thereof)…
• Data Gap Analysis — Data Gap Analysis is the exercise of reconciling a source system's native data shape against a target system's required template, producing a structured two-axis output:
• Data governance tiering — Data governance tiering is the discipline of classifying every table in a data warehouse into a small number of quality tiers — typically three…
• Data integrity checker — A data integrity checker is a routine (typically SQL, sometimes custom code) that compares the output of a data pipeline against an independent source of truth and reports…
• Data Lakehouse — A data lakehouse is a data-platform architectural class that combines:
• Data lineage — Data lineage is the graph of relationships between data assets that tracks "where did this data come from" and "where does this data flow to" — source → sink relationships…
• Data Mesh — Data mesh is an architectural approach to organisation-wide data sharing in which data products are owned and exposed by domain teams (R&D, After-Sales, Marketing, etc.),…
• Data minimization — Data minimization is the discipline of transmitting, processing, and storing only the data actually needed for the immediate task — no surrounding context,…
• Data model mismatch (OLTP vs OLAP) — Data model mismatch names the structural observation that OLTP and OLAP workloads want different shapes of the same data,…
• Data-oriented service mesh — A data-oriented service mesh is a single GraphQL-shaped interface for accessing and interacting with any backing data source within an organization,…
• Data parallelism — Data parallelism (DP) is the simplest axis of distributed training: replicate the model across N workers, shard the mini-batch into N pieces,…
• Data pipeline grain — Data pipeline grain is the time (or other dimensional) resolution at which a pipeline processes its inputs and emits its outputs.…
• Data Plane Atomicity — Data Plane Atomicity is Redpanda's design tenet for BYOC cloud deployments: no deployment can bring down any other deployment, including a control-plane outage,…
• Data–policy separation — Data–policy separation is the architectural discipline of keeping authorization policy logic completely independent from authorization data loading:
• Data pre-processing vs data transformation split — Data pre-processing and data transformation are two distinct stages of feature-engineering pipelines, with different objectives, tooling, and scaling characteristics.…
• Data-quality checksum comparison — Data-quality checksum comparison is the canonical correctness primitive for verifying that two parallel sources of the same logical data agree: compute a row count and a checksum…
• Data upload format validation — Data upload format validation is the pre-upload check that a data file about to be sent to a downstream system (typically a third-party SaaS) actually conforms to the downstream…
• Database-agnostic abstraction — A database-agnostic abstraction is an interface that lets applications describe their data problem — access patterns, consistency needs,…
• Database availability attainment — Database availability attainment is the measurement substrate that tracks reliability per individual database rather than as a fleet aggregate,…
• Database branching — Database branching is the workflow primitive of creating an isolated sandbox copy of a production database's schema (and, on some platforms,…
• Database development compensating layer — The database development compensating layer is the stack of workarounds that the software industry built over twenty years (2003–2023) to compensate for the absence…
• Database fleet standardisation — Database fleet standardisation is the discipline of collapsing per-team, per-database operational practice — thresholds, dashboards, ad-hoc scripts,…
• Database permission migration risk — A class of incident where a correct, defensive permission change on a database causes a downstream consumer to break because the consumer had an implicit assumption about what its…
• Database startup time SLI — Database startup time is a Service Level Indicator (SLI) specific to serverless / scale-to-zero database services: how long it takes a suspended database to become available…
• Database throttler — A database throttler is a service or component that pushes back on an incoming flow of database operations so that the database system stays healthy under load.…
• Database transaction — A database transaction is a sequence of SQL statements executed as a single atomic unit of work, bracketed by begin; at the start and either commit;…
• DBA as design collaborator — DBA as design collaborator names the role evolution of the database administrator from synchronous gatekeeper of production-schema safety to asynchronous collaborator on schema…
• DBA as forced gatekeeper — DBA as forced gatekeeper is Shlomi Noach's 2021 structural observation that at a growing company, the DBA's return to gatekeeper status is not a cultural regression…
• DBA as platform engineer — DBA as platform engineer is the role evolution where the database administrator shifts from provisioning/ops gatekeeper to platform architect who designs the structural framework…
• DBSQL caching tiers — A two-tier cache hierarchy in a Databricks SQL warehouse: a disk cache (warehouse-local SSD holding hot Parquet files / data blocks) and a query result cache (QRC) (full query…
• DCQCN (Data Center Quantized Congestion Notification) — DCQCN is the de-facto standard end-to-end congestion-control algorithm for RoCE networks. It combines ECN (Explicit Congestion Notification) marking at switches with a rate-control…
• DDSketch error-bounded percentile — A DDSketch (arXiv:1908.10693) is a streaming quantile sketch that provides relative-error guaranteed percentiles over a stream of numeric values without retaining the raw values.…
• Dead-code elimination — Dead-code elimination (DCE) is a linker / compiler pass that drops symbols the reachability analysis proves no entry-path reaches.…
• Dead man's switch — A dead man's switch (DMS) is a monitoring primitive where a continuously-firing heartbeat is delivered to an external channel,…
• Dead tuples and VACUUM — A dead tuple is a Postgres row version that has been logically deleted or superseded by an update but not yet physically removed.…
• Deadlock vs lock contention — Deadlock and lock contention can look identical to an outside observer watching a process hang, but they are different failure modes with different fixes:
• Decentralized development of a central schema — Decentralized development of a central schema is the problem of letting many teams contribute to one shared API schema — without forcing every change to flow through a central…
• Decision-support vs audit query — A decision-support query is an analytical query whose answer only needs to be good enough to drive the same decision as the exact answer.…
• Declarative agent definition — A framework design principle where developers describe what an agent knows (model, skills, sandbox, instructions) rather than scripting what it does (no explicit orchestration…
• Declarative lifecycle API — A declarative lifecycle API is a contribution-surface design where a component author describes a fixed set of named phases (e.g.…
• Declarative load-test API — A declarative load-test API is one where the client describes the desired end state of a load-test run — target load (e.g., orders-per-minute), duration, ramp-up + plateau times,…
• Declarative quota rule — A declarative quota rule expresses the quota's intent as structured data — subject + scope + predicate + limit + action — so that a platform can read, validate, reason about,…
• Declarative vs Imperative Stream API — The declarative vs imperative stream API tradeoff is the choice between expressing a streaming computation as a query (SQL / relational / planner-optimised) versus as an explicit…
• Declared crawler — A declared crawler is an automated web client whose operator has published:
• Decouple reasoning from structured output — Decouple reasoning from structured output is the two-pass LLM design in which:
• Dedicated-but-managed infrastructure — Dedicated-but-managed infrastructure is the organisational pattern where a consuming team (e.g., observability, security,…
• Deep-link to screen-config resolution — In a server-driven UI system, a deep-link (URL that opens a specific app screen) is resolved into a screen configuration through an indirection middleware rather than a hardcoded…
• Deep Reinforcement Learning congestion control — Deep Reinforcement Learning (DRL) congestion control models the sender's pacing decision as a policy learned from experience, rather than hand-tuning heuristics like CUBIC or BBR.…
• Default Access Retention — Default Access Retention (DAR) is Yelp's named middle-ground retention primitive between deletion-based retention (data is gone) and cold-tier-by-default (data remains freely…
• Default-closed table allowlist — A default-closed table allowlist is a data-platform governance posture in which tables are inaccessible for querying until they have been reviewed,…
• Defense in Depth — Defense in depth is the security posture of stacking independent protective layers such that no single compromised or misconfigured layer exposes the whole system.…
• Deferred join — A deferred join is a SQL rewrite technique that replaces a wide SELECT col1, col2, … FROM t ORDER BY k LIMIT N OFFSET M (which must hydrate M+N full rows from the clustered index)…
• Delayed symbolization — Delayed symbolization is the practice of deferring the address-to-symbol resolution step until after the profile collection step on the profiled host has completed…
• DELETE-LIMIT asymptote — The DELETE-LIMIT asymptote is a latency-degradation shape seen on recurring DELETE … WHERE LIMIT jobs on an unindexed predicate.…
• Deletion vector — A deletion vector is a compact file-level representation of "these row positions in this data file are logically deleted",…
• Delta chain replay — Delta chain replay is the read-path mechanism a compute-storage-separated database uses to materialise a page on demand: find the most recent full page image of that page…
• Delta Change Data Feed — Delta Change Data Feed (CDF) is a Delta Lake capability that exposes the row-level change events — inserts, updates, deletes — that occurred between two versions of a Delta table,…
• HTTP delta compression (dcb / dcz) — HTTP delta compression is the wire-format layer that carries the output of shared-dictionary compression over HTTP. Standardised by RFC 9842,…
• Delta encoding — Delta encoding is a reversible transform that replaces each value in a sequence with the difference from its predecessor: x[i] → x[i] - x[i-1].…
• Demand-driven replication — Demand-driven replication is the policy of materialising a replica of an object in a region only when that region actually reads the object,…
• Denoise → encode → synthesize — Denoise → encode → synthesize is the three-stage encoding-pipeline shape used by AV1 Film Grain Synthesis. Rather than passing the source video directly into the codec,…
• Dense semantic retrieval — Dense semantic retrieval (also embedding-based retrieval, EBR) is the class of information-retrieval approaches that encode a query and its candidate documents into dense vectors…
• Dense Transformer — Dense Transformer is the frontier-LLM architectural shape in which the model is a single transformer stack — every token passes through every parameter at every layer…
• Deny-by-default tool policy — A deny-by-default tool policy is the architectural posture for AI agent governance in which tools are denied unless explicitly allowed…
• Dependency count by language ecosystem — Dependency count by language ecosystem is the empirical observation that the number of transitive dependencies per application varies by 1–2 orders of magnitude across language…
• Dependency on self — A circular-dependency failure mode where an organisation operates the product it itself depends on — "X runs on X" — so that an outage of the product takes out the very tools…
• Dependent-destroy cascade risk — Dependent-destroy cascade risk is the unbounded-work failure mode of ActiveRecord's hasmany :children, dependent: :destroy option when child-row count grows with application age.…
• Deploy frequency vs caching — Deploy frequency vs caching is the structural tension between two second-order goods: higher deploy frequency (faster iteration, faster product velocity,…
• Deploy-less operational lever — A deploy-less operational lever is any production- behaviour control whose invocation is a config flip (feature-flag toggle, parameter change,…
• Deploy-marker annotation — A deploy-marker annotation is a vertical event overlay on a time-series graph marking the moment a deploy (schema change, code release) took effect.…
• Deploy request — A deploy request is PlanetScale's first-class primitive for proposing a schema change against a target database. It is the database-schema analog of a pull request: an engineer…
• Deployment transparency evidence — Deployment transparency evidence is a public, reproducible audit artifact that documents a specific infrastructure deployment event was provisioned under the claimed process.…
• Derived-key aggregation — Derived-key aggregation is the logging discipline of counting cryptographic operations on a KDF-derived child key against the parent keyset's name rather than against the child…
• Derived subtree — A derived subtree is a subtree of a document tree whose structure and property values are not authored directly but computed from other sources of truth…
• Descent-into-madness debugging — A named phase in debugging hard production bugs: the phase after all your working models have been invalidated by evidence, where you start wild-guessing,…
• Design Away Invalid States — Design away invalid states is the architectural principle that invalid states should be unrepresentable — at the protocol level, the data model level,…
• Design-to-code traceability — Design-to-code traceability is the architectural property of maintaining a visible, verifiable link between decisions made during design review (threat models, API contracts,…
• Design token as named reference — A design token serialised as named reference is a wire-format representation of a design-system token (colour, icon, gradient, shadow,…
• Design-token cross-platform theming — Design-token cross-platform theming is the approach of defining styling variables (colours, font sizes, border widths, spacing,…
• Destroy-async (Rails 6.1+) — dependent: :destroyasync is a Rails 6.1+ association option that triggers cascade deletion of child records via a background ActiveJob job rather than inside the parent's destroy…
• Destroy vs delete in ActiveRecord — destroy and delete are ActiveRecord's two row-removal primitives on ActiveRecord::Base (single record) and ActiveRecord::Relation (collection).…
• Destructive-automation blast radius — Destructive-automation blast radius is the scope of damage an automated destructive operation (delete, drop, terminate, detach,…
• Destructive-query protection — Destructive-query protection is a server-side static check on a SQL string that blocks structurally-dangerous statements before execution,…
• Detection-in-depth — Detection-in-depth is the security posture of layering continuous detection on top of preventive controls, so that when (not if) prevention misses a class of bug,…
• Determinantal Point Process (DPP) — A Determinantal Point Process (DPP) is a probability distribution over subsets of a ground set that assigns higher probability to subsets of items that are both high-quality…
• Deterministic Simulation — Deterministic simulation is a testing discipline where the entire system under test runs inside a custom executor/scheduler that eliminates every source of non-determinism it…
• Deterministic state machine for lifecycle — A deterministic state machine for lifecycle is the pattern of modelling the lifecycle of a stateful resource (a query, a request, a workflow, a connection,…
• Developer platform over bare substrate — A developer platform over bare substrate is the architectural / commercial posture where a managed-service vendor takes a powerful but operationally complex open-source primitive…
• Developer schema workarounds — Developer schema workarounds are the four anti-patterns Shlomi Noach observed (PlanetScale, 2021) developers reach for when the operational friction of shipping a schema change…
• Developer velocity — Developer velocity is the composite metric tracking how quickly and confidently an engineering org can ship changes. It spans multiple sub-metrics depending on how the org defines…
• Developers want LLMs, not GPUs — The demand-side observation that the developer audience for "AI-enabling their app" overwhelmingly wants an LLM API (token in, text out) — not a GPU, not a model, not a framework,…
• Device certification program — A device certification program is a structured process by which a streaming service validates that a candidate playback device — TV, set-top box, streaming stick, game console,…
• Device-to-host (D2H) transfer — A device-to-host transfer (D2H copy) is the operation of moving bytes from GPU memory (VRAM) to CPU-accessible host memory (RAM).…
• Device trust — Device trust is the security posture where a specific piece of hardware — typically an employer-managed laptop — is provisioned with a short-lived cryptographic credential…
• Dewey Decimal Classification — Dewey Decimal Classification (DDC) is the universal library classification system that organises knowledge into ten primary classes (000–900) with hierarchical decimal…
• Diff commutativity check — The diff commutativity check is the conflict-detection test at the heart of schema three-way merge: given two concurrent semantic schema diffs diff1 and diff2,…
• Diff noise filtering — Diff noise filtering is the preprocessing step that strips non-reviewable or auto-generated content from a code diff before any AI reviewer sees it.…
• Diff-overlap auto-adaptation — Diff-overlap auto-adaptation is the schema three-way merge rule that treats identical sub-diffs shared between two concurrent branches as benign and auto-resolves them at merge…
• Diff profile regression analysis — Diff profile regression analysis is the incident-response technique of comparing two continuous profiles — one from before a regression,…
• Diff sketch — A diff sketch is a machine-readable summary of a code change — what classes / methods / fields / calls were introduced, removed,…
• Differential fuzzing — Differential fuzzing is the testing technique of feeding the same randomly/fuzz-generated input to two (or more) implementations of a specification and treating any divergence…
• Differential privacy — Differential privacy (DP) is a mathematical guarantee that the output of a computation is statistically insensitive to whether any single individual's data was included or not.…
• Digital sovereignty — Digital sovereignty is "managing digital dependencies — deciding how data, technologies, and infrastructure are used, and reducing the risk of loss of access, control,…
• Digital-twin backtesting — Digital-twin backtesting is the technique of running counterfactual or evaluation workloads against a high-fidelity simulated replica of a production system,…
• Direct Preference Optimization (DPO) — Direct Preference Optimization (DPO) is a post-training method that trains an LLM on pairs of (preferred response, rejected response) examples by optimising a loss derived…
• Direction-agnostic edge ID — A direction-agnostic edge ID is an edge identifier constructed deterministically from the pair of endpoint node IDs in a way that produces the same identifier regardless of which…
• Dirty read — A dirty read is a read-anomaly class that occurs when one transaction observes another transaction's uncommitted writes. If the writing transaction later rolls back,…
• Disaster recovery tiers (backup / pilot light / warm standby / active-active) — The canonical AWS-lineage disaster-recovery ladder: four tiers ordered by cost, complexity, and recovery time. Picking a tier is a choice of where to trade ongoing cost for lower…
• Discharge token — A discharge token is the companion Macaroon that satisfies a third-party caveat. The permissions Macaroon says "valid only if accompanied by a discharge from authority X";…
• Discrete Event Simulation — Discrete Event Simulation (DES) models a stochastic system as a sequence of discrete events at discrete time steps, where at each step the system state evolves deterministically…
• Discrete event vs heartbeat attribution — Two structural shapes for maintaining an owner mapping that changes over time in a distributed system:
• Disk block size alignment — On-disk data structures achieve their best performance when the unit of the data structure (node, page, block) matches the unit of the storage device's I/O.…
• Disk throughput bottleneck — Disk throughput bottleneck — the regime where a system's performance ceiling is set by how many bytes per second the underlying storage device can read or write, not CPU,…
• DiskANN index — A DiskANN index is an SSD-resident graph ANN index based on the Vamana graph-construction algorithm (Microsoft Research, NeurIPS 2019).…
• Distributed isolation coupling cost — Distributed isolation coupling cost is the thesis that isolation-level strictness in a distributed database is not just a concurrency-correctness knob — it is a coupling lever.…
• Distributed log — A distributed log is an ordered, append-only sequence of records distributed across multiple nodes, addressed by a monotonically increasing offset, with:
• Distributed monolith — A distributed monolith is the anti-pattern of having successfully decomposed a monolith into many microservices — but keeping the underlying coupling intact.…
• Distributed transactions — Distributed transactions are atomic operations that span multiple database rows, shards, or nodes — committing all parts or none, across machine boundaries.…
• Distribution quality vs p99 tail — Distribution quality is a measurement-philosophy axis for latency / performance metrics: do you optimise to maximise the share of observations in a "fast / instant" bucket (i.e.…
• Diurnal autoscaling risk — Diurnal autoscaling risk is the structural vulnerability that arises when a fleet scales on a predictable 24-hour pattern — scale up before a known peak-traffic window,…
• Diurnal traffic pattern — The diurnal traffic pattern is the day/night rhythm visible in any aggregate measurement of human-driven Internet traffic at country / regional / metropolitan scale: traffic rises…
• Diversity via beam width and temperature — In generative retrieval systems, beam width and temperature are runtime knobs that let one model serve multiple surfaces with different precision-vs-exploration trade-offs without…
• DNS message ordering — DNS message ordering is the question of what order records must appear in within a DNS response message's Answer section…
• DNS outage recovery — DNS outage recovery is the operational category of restoring service after DNS entries themselves have been lost, corrupted,…
• DNS rebinding SSRF risk — DNS rebinding (as an SSRF risk) is the structural observation that the IP address a hostname resolves to at validation time is not guaranteed to be the same IP it resolves…
• DNS request amplification via retries — When DNS queries are routed through multiple layers of resolvers (e.g. application → local per-host resolver → central cluster resolver → upstream authoritative),…
• DNS resolver caching — Recursive DNS resolvers (such as Unbound) cache the results of DNS queries with TTL-bounded lifetimes so repeated lookups for the same name don't hit the upstream on every call.…
• DNS reverse lookup (PTR record) — Reverse DNS is the DNS operation of looking up a hostname given an IP address — the inverse of normal forward resolution.…
• DNS SERVFAIL response — SERVFAIL (RCODE 2) is the generic "server failure" response code a DNS server returns when some error occurred during resolution: upstream timeout, policy rejection,…
• DNSSEC chain of trust — The DNSSEC chain of trust is the transitive authority chain from the DNS root zone down to an individual signed DNS record.…
• DNSSEC — DNSSEC (Domain Name System Security Extensions; RFC 4033 + 4034 + 4035) adds cryptographic authentication to DNS. When a zone is signed with DNSSEC,…
• DO-to-Container cross-region RTT — DO-to-Container cross-region RTT is the latency additivity hazard introduced by DO-enabled Containers when the Durable Object is placed near the user but the connected Container…
• Document sharding — Document sharding is the horizontal-scale-out primitive used in Meta's SilverTorch retrieval substrate when "the neural network exceeds a single host's capacity" (Source:…
• Document storage compression — Document storage compression is the per-block compression applied by a database's storage engine before writing pages to disk;…
• DOM context injection for LLM — DOM context injection for LLM is the technique of capturing the actual rendered DOM of a UI component at test-run time and feeding it to an LLM as disambiguating context…
• DOM node count — DOM node count is the total number of elements the browser has to hold + style + lay out + paint for a given page. At modern browser implementations each node is hundreds of bytes…
• Domain adaptive learning — Domain Adaptive Learning (DAL) is a subset of transfer learning in which the task stays the same across two settings but the input distribution changes — for example,…
• Domain-knowledge redundant condition — A domain-knowledge redundant condition is a redundant-condition variant whose redundancy depends not on logic but on an application-level invariant the database cannot prove.…
• Domain model — A domain model — in the data-integration / knowledge-graph sense Netflix UDA uses — is a controlled vocabulary that defines:
• DORA metrics — DORA metrics — four software-delivery performance indicators established by the DevOps Research and Assessment program led by Dr.…
• Double-checked locking — Double-checked locking is a concurrency idiom that avoids duplicating expensive work when multiple threads / requests / processes compete to do the same thing. The structure is:
• Double-click-selectable identifier — A double-click-selectable identifier is an API identifier whose character set is narrow enough that browsers and terminal emulators treat the whole string as a single word when…
• Double NAT — Double NAT is the configuration where a packet undergoes two stateful Network Address Translation stages on the same host before leaving it (or on the return path, in reverse).…
• Downgrade attack — A downgrade attack is an active-adversary manipulation of a cryptographic protocol negotiation that forces the two endpoints to select a weaker algorithm (or a lower protocol…
• DR configuration translation — DR configuration translation is the problem — and the mechanisms that solve it — of rewriting application configuration at failover time because restored resources have new…
• Draft schema field — A draft schema field is a GraphQL field marked — typically by a custom directive — as "not ready for production". The persist step that registers queries into the persisted-queries…
• Drafter-expert split — Drafter-expert split is the architectural primitive under both speculative decoding and cascades: two models are co-hosted on the same LLM-serving stack — a small,…
• Drill muscle memory — An operational-readiness discipline where an organisation regularly exercises its emergency procedures — break-glass authorisation, backup communication channels,…
• Dual-architecture compatibility — Dual-architecture compatibility is the discipline of keeping a single codebase buildable and testable under both the old and new versions of an underlying runtime or renderer,…
• Dual clustering for document intelligence — Dual clustering is a two-lens document analysis pattern in which every document is clustered simultaneously along two complementary axes — semantic (meaning,…
• Dual-owner metric governance — Dual-owner metric governance is an ownership model where every governed metric has two mandatory, team-based owners with distinct responsibilities:
• Dual-write branch migration — Dual-write branch migration is the technique of migrating a sharded database to a new cluster (usually with a different shard count) by provisioning the destination cluster empty,…
• Dummy-write heartbeat kludge — The dummy-write heartbeat kludge is the industry workaround for runaway WAL growth on a Postgres primary feeding logical replication: a scheduled job periodically writes a dummy…
• Durability as use-case dependent — Durability as use-case dependent is the architectural stance that the durability requirement of a consensus system — how many acknowledgments, from which nodes,…
• Durability vs. consistency guarantee — Durability and consistency are two orthogonal guarantees a replicated system can offer. A system can provide one without the other.…
• Durable execution — Durable execution is the property of a long-running computation (multi-minute LLM loops, multi-hour CI pipelines, multi-day workflows) that it survives any interruption of its host…
• Durable request — A durable request is one that has met the system's durability requirement — enough followers responsible for the leader's durability have acknowledged receipt that the system can…
• Durable state as URL — A workload-state design in which the authoritative, durable identity of a workload's state is a URL (or a small handle that resolves to a URL) into an object store.…
• Durable Streams — An append-only, immutable event log applied to LLM agent execution history. Every event — prompt, tool response, model choice — is recorded as an unchangeable ledger entry.…
• Durable vs ephemeral sandbox — A design-axis distinction for VM/container sandboxes hosting coding agents: does the sandbox persist across agent sessions (durable) or vanish at session end (ephemeral)?
• DWARF debug info — DWARF (Debugging With Attributed Record Formats) is the standard debug-information format emitted by Unix toolchains (gcc, clang) into ELF binaries.…
• Dynamic backend fallback — Dynamic backend fallback is a resilience pattern where a system starts a session using a preferred backend and, on mid-session failure of that backend,…
• Dynamic partition splitting — Dynamic partition splitting is the runtime remediation pattern of detecting individual partitions that have grown beyond the storage engine's healthy operating envelope…
• Dynamic sampling rate tuning — Dynamic sampling rate tuning is a feedback-control loop that adjusts a sampling profiler's run probability (or rate) per-service, per-host, or per-profiler, on a regular cadence,…
• Dynamic schema — field names as data — A document-database schema where field names are not pre-defined constants but carry information — e.g. the day-of-month, a user-ID,…
• Dynamic sharding — Dynamic sharding treats a service's shard assignment as state owned by a controller, continuously and asynchronously updated in response to pod health, load, termination notices,…
• Dynamic workflow configuration — Dynamic workflow configuration is the design choice of exposing per-workflow runtime knobs through the control plane (workflow create/update API) rather than through process-level…
• DynamoDB Streams — DynamoDB Streams is Amazon DynamoDB's built-in change-data-capture (CDC) feed. When enabled on a table, every insert / modify / delete of an item produces a corresponding stream…
• Early-exit logits — Early-exit logits are the vocabulary-space logit vectors obtained by applying the transformer's final projection matrix (LM-head Wlm) to the hidden state of an intermediate…
• Early fraud warning — An early fraud warning (EFW) is a notification issued by a card network (Visa / Mastercard / etc.) that a transaction is likely fraudulent, issued ahead of a formal chargeback.…
• Early Fusion (Multimodal) — Early fusion is the multimodal-ML strategy of concatenating raw (or minimally-encoded) modality inputs into a single feature vector before training,…
• Earthquake Early Warning (EEW) — Earthquake Early Warning (EEW) is the class of real-time systems that detect the onset of a seismic event near its source, estimate its parameters (epicentre, magnitude,…
• EAV schema for LLM consumption — An Entity-Attribute-Value (EAV) schema stores heterogeneous attributes of entities as (entityid, attributename, value, …) rows — one row per attribute per entity…
• eBPF profiling — eBPF profiling is the practice of collecting profile samples (CPU stacks, memory allocations, latency buckets, off-CPU time,…
• eBPF verifier — The eBPF verifier is the static-analysis pass the Linux kernel runs on every eBPF program before it's allowed to load. It proves that the program is safe to execute in kernel…
• EBS IOPS burst bucket — An EBS IOPS burst bucket is a credit-bucket rate limiter on Amazon EBS volumes: while the volume runs below its configured IOPS cap,…
• EC2 launch failure mode — The EC2 launch failure mode is the operational fault class in which running EC2 instances continue to serve traffic normally but RunInstances (or equivalent launch API calls)…
• ECMP (Equal-Cost Multipath) — ECMP (Equal-Cost Multipath) is the standard Ethernet/IP load-balancing primitive for Clos / leaf-spine fabrics: when multiple next-hops have the same routing cost to reach…
• Edge-to-cloud data flywheel — An edge-to-cloud data flywheel is an ML-platform pattern where a deployed fleet of edge devices continuously feeds a cloud data-management + training platform,…
• Edge filtering — Edge filtering is the pipeline design move of dropping / matching events at the producer (the host Agent, sidecar, or data plane closest to the event source),…
• Edge-to-origin database latency — Edge-to-origin database latency is the extra per-query latency that accrues when a serverless / edge compute unit runs far from a centrally-located transactional database…
• Effective batch size — Effective batch size is the actual batch size a streaming- broker producer achieves in production, as distinct from the configured ceiling (batch.size) the operator sets.…
• Egress Cost — Egress cost is the per-byte charge a cloud provider levies when data leaves a region, a cloud, or (in some cases) an availability zone.…
• Egress SNI Filtering — Egress SNI filtering is the pattern of enforcing an outbound allow-list based on hostnames matched against the TLS ClientHello's Server Name Indication (SNI) field rather than…
• Elasticity — Elasticity is the property that a service's capacity and performance expand and contract to customer demand without requiring the customer to forecast, provision,…
• Elasticsearch compatibility mode — Elasticsearch compatibility mode is a mechanism Elastic ships in its 7.17 High Level REST Client (HLRC) that lets the old client talk to the new cluster during a major-version…
• ElasticsearchDataSet (EDS) — ElasticsearchDataSet (EDS) is the Kubernetes Custom Resource Definition used by Zalando's es-operator to declaratively describe an Elasticsearch cluster.…
• Election-related Internet shutdown — An election-related Internet shutdown is a government-directed shutdown whose policy trigger is a national election. Authorities typically cite "misinformation, disinformation,…
• Elector — An elector is an agent permitted to run the leadership-change protocol (detect failures, revoke the prior leader, establish a new one).…
• Elicitation gate — An elicitation gate is the agent-architecture mechanism by which an agentic system solicits approval, preferences, or missing information from the user or another system…
• Elm architecture — The Elm architecture (often abbreviated TEA — The Elm Architecture) is a UI programming pattern originated by the Elm programming language that decomposes any interactive…
• ELT vs ETL — ETL (Extract → Transform → Load) transforms data before loading it into the warehouse, usually in an external worker. ELT (Extract → Load → Transform) lands raw data into…
• Email as agent interface — Email as agent interface is the framing that treats an email inbox not as a transactional notification sink but as a bidirectional channel between a user and an autonomous agent…
• Embedded analytics — Embedded analytics is the pattern of serving BI dashboards and reports inside a host application's user experience — typically via an iframe or SDK widget…
• Embedded OPA library in proxy — Running Open Policy Agent as a library inside the ingress / proxy process, not as a sidecar, not as an external deployment, not as a separate HTTP service.…
• Embedded routing in IP address — Embedded routing in IP address is the addressing design where routing information is encoded directly into the IP address itself,…
• Embedded SRE team — An Embedded SRE team is a full SRE team that sits within a product area but reports up the SRE department chain. It is distinct from the lone-SRE-per-team anti-pattern (one…
• Embedded workflow engine — An embedded workflow engine is a durable-execution engine delivered as a library dependency that runs inside the host service's process,…
• Embedding black-box debugging — Embedding black-box debugging names the operational failure mode where a vector-database- backed retrieval pipeline returns the wrong chunk and the debugging path to understand why…
• Embedding Collection — An embedding collection is the organizational unit of a vector database: a named, schema-pinned bucket of vector embeddings that share
• Embedding-dimension diminishing returns — Embedding-dimension diminishing returns is the observation that increasing a vector-embedding's dimensionality past a certain threshold — empirically ~1,536 dimensions…
• Embedding in context — Embedding in context is an inference-time deployment rule: when using a Transformer encoder that produces contextualised embeddings over a sequence,…
• Embedding signal dilution — Embedding signal dilution is the empirical observation that encoding a long text into a single dense vector degrades retrieval quality because the relevant signal…
• Embedding version skew — Embedding version skew is the production failure mode where an ML system's query-side embeddings and item-side embeddings are computed by different model checkpoints because…
• Emergency failure probe — An emergency failure probe is an out-of-cadence health check fired when holistic detection yields an asymmetric signal — one observer reports the primary down while others…
• Empty host — Empty host is the operational primitive of a fully-empty physical server with nothing scheduled on it, deliberately preserved by the cluster scheduler for:
• Encoded domain expertise — Encoded domain expertise — what Meta calls "skills" on the Capacity Efficiency Platform — is the reusable, composable artifact form of a senior engineer's reasoning playbook…
• End-to-end encryption (E2EE) — End-to-end encryption (E2EE) is the property that a message is encrypted on the sender's device, decrypted only on the recipient's device,…
• End-to-end test probe — An end-to-end test probe is a small set of browser-driving end-to-end test scenarios repurposed as a periodic external monitor of a live production system.…
• Enhanced ECMP with Queue-Pair scaling — Enhanced ECMP (E-ECMP) with Queue-Pair scaling is a routing + collective-library co-design technique for getting good load balance out of an Ethernet training fabric without giving…
• Enrichment execution engine — An enrichment execution engine is the shared runtime substrate that turns raw events into enriched records based on configuration — "reading configuration,…
• Entitlement / budget / quota integration — A three-layer governance hierarchy linking dollars to enforced resource limits:
• Entity-based page composition — Entity-based page composition is a page-assembly model in which the request-time structure of a page is a tree of typed Entities,…
• Entity ID convention — An entity ID convention is a project-wide rule that every identifier exposed on an API boundary follows the same structural pattern,…
• Entity resolution — Entity resolution (ER) — also called record linkage, deduplication, or identity resolution — is the problem of deciding whether two or more records,…
• Entity URI namespace — An entity URI namespace is a catalog-design primitive that assigns every entity a globally-unique URI of the form:
• Entropy repair — Entropy repair is the family of mechanisms a distributed system uses to discover and remediate inconsistencies that crept in via partial failures.…
• Envelope Encryption — Envelope encryption is a multi-level key-hierarchy scheme for encrypting data at rest. Data is encrypted with unique per-segment Data Encryption Keys (DEKs),…
• Envelope wrap-and-unwrap metadata routing — Envelope wrap-and-unwrap metadata routing is a wire-format technique for threading routing metadata (tenant ID, version, account key, agent ID,…
• Environmental parity — Environmental parity is the principle, applied during a platform migration, that the new runtime environment must be indistinguishable from the old one from user code's point…
• Ephemeral credential injection — A security pattern for multi-user agent sandboxes where user credentials are injected into the sandbox environment only for the duration of a single command execution,…
• Ephemeral credentials — Ephemeral credentials are credentials — keys, tokens, certificates, or passwords — that are generated on demand, used for a short lifetime, and discarded.…
• Ephemeral dev environment — A development environment whose lifetime is bounded by a single session (or task, or agent run) rather than existing as a long-lived personal resource.…
• Ephemeral local disk vs EBS — A classic AWS-era storage tradeoff for stateful services: run on network-attached block storage (EBS) — durable, movable between instances, slower,…
• Ephemeral preview URL — An ephemeral preview URL is a publicly-reachable URL — typically on a platform-owned subdomain — that points at a bound port inside an ephemeral development or session environment…
• Ephemeral sensitive data — Ephemeral sensitive data is information that is (a) provided by the user at runtime for a specific interaction, (b) sensitive enough that durable storage materially raises…
• Ephemeral storage cross-zone drift — Ephemeral storage cross-zone drift is the property that a StatefulSet pod on ephemeral storage — emptyDir, local-instance SSDs,…
• Epoch-based distributed garbage collection — Epoch-based distributed garbage collection is a reclamation technique in which:
• Erasure coding — Erasure coding is a redundancy scheme that encodes data into more pieces than are needed to read it, so that the data survives the loss of any fixed number of pieces.…
• Errant GTID tracking — An errant GTID (Global Transaction ID) is a transaction that exists on a replica but not on the primary of a MySQL replication topology…
• Error action-taker classification — Error action-taker classification is a decision framework that classifies each failure mode by who can actually take recovery action on it.…
• Error budget — The error budget is the complement of an SLO: if the SLO target is 99.9% over a 28-day window, the error budget is the 0.1% of traffic the service is allowed to fail.…
• Error propagation — Error propagation is the deterministic downstream consequence of an upstream-stage mistake in a multi-stage pipeline, when no later stage has the ability to re-examine the original…
• Escrow VPC for LLM serving — An escrow VPC for LLM serving is a deployment shape in which a third-party model provider's proprietary model weights are hosted in a customer-controlled cloud environment under…
• Estimated Time of Arrival (ETA) — Estimated Time of Arrival (ETA) is the predicted clock time at which a trip (vehicle, delivery, transit, courier) will reach its destination, given the origin, the route,…
• ETag conditional polling — ETag conditional polling is a cache-coherence protocol in which a client on each poll sends its last-seen ETag (an opaque fingerprint of the resource) and the server replies:
• Evaluation label — An evaluation label is the unit of offline agent evaluation. It has two parts, and the key design move is that only one of them is visible to the agent under test:
• Event-driven architecture — Event-driven architecture (EDA) is a software-system style in which services communicate asynchronously by publishing and subscribing to events on a shared bus…
• Event-log-based counter — An event-log-based counter stores each counter mutation as an individual event in an event store, and computes the counter's value by aggregating events.…
• Event-loop blocking in single-threaded services — Event-loop blocking is the failure mode where a single-threaded runtime (Node.js, V8 isolates, LuaJIT / OpenResty, CPython's GIL-bound loop,…
• Competing sources of truth (event stream) — Competing sources of truth is the anti-pattern where multiple independent systems each claim to hold the authoritative version of a domain entity (product, order, customer),…
• Event-triggering orchestration — Event-triggering orchestration is the architectural stance that workflow-to-workflow integration across teams should happen via a shared event / signal protocol…
• Eventual consistency too slow for allocation — Eventual consistency too slow for allocation is the failure mode where an eventually-consistent store's convergence time exceeds the rate at which an exclusive resource is claimed…
• Eventual consistency — Eventual consistency is a liveness guarantee: if no new updates are made to a shared value, all observers will eventually converge on the same read.…
• Eventual durability — A durability class in which a write is acknowledged to the application before it reaches the authoritative durability tier (typically object storage).…
• Evidence library for disputes — An evidence library for disputes is a merchant-curated document corpus — terms and conditions, return policies, service agreements,…
• Evolutionary database design — Evolutionary database design is the discipline of treating database schemas as first-class artefacts that evolve incrementally alongside application code,…
• Exactly-once signal trigger — An exactly-once signal trigger is a signal-driven workflow invocation where the orchestrator guarantees the subscribing workflow runs exactly once per matching signal…
• Example-based test constant-input antipattern — The example-based test constant-input antipattern is the failure mode where a unit test hand-types specific constant values as inputs,…
• Execution ladder — The execution ladder is a tiered-environment framing for agent code execution in which each tier adds capability on top of the previous one,…
• Exit code semantics — Exit code semantics is the commitment that a CLI's exit codes and error shapes are documented, consistent, and load-bearing for agent control flow…
• Experience (Zalando Interface Framework) — An Experience in Zalando's Interface Framework is a named bundle of (policies, selection rules) resolved once per request from the customer's context and then applied to every…
• Experimentation culture — Experimentation culture is the organisational state in which data-driven decisions via controlled experiments are the default,…
• Experimentation Evolution Model (Fabijan et al.) — The Experimentation Evolution Model is a four-stage maturity framework for online controlled experimentation in software organisations, introduced in Fabijan et al.,…
• Expert parallelism (MoE serving) — Expert parallelism is a multi-GPU model-sharding strategy specific to Mixture-of-Experts (MoE) models: different experts (sub-networks activated by a learned router per-token) live…
• Explainable AI decision — An explainable AI decision is one whose output is accompanied by a first-class, queryable justification — not a post-hoc reconstructed one…
• Explicit graphics state — Explicit graphics state is an API-design choice where every draw call specifies the full state it needs — vertex buffers, textures, materials, pipeline,…
• Explicit state over implicit sync — Explicit state over implicit sync is the storage- schema discipline of representing cross-platform (or cross-client, or cross-device) state as independent explicit values per…
• Explicit teardown on completion — Explicit teardown on completion is the reliability property where reaching a terminal state of a state machine (e.g. Finished, Cancelled,…
• Exploit chain construction — Exploit chain construction is the process of taking several small, individually-low-severity attack primitives and stitching them together into a single working exploit.…
• Exploration-exploitation tradeoff in agent search — The exploration-exploitation tradeoff in LLM-agent search is the per-rollout decision: should the agent refine a known-promising candidate (exploit) or try a risky-but- informative…
• Exponential backoff with jitter — Exponential backoff with jitter is the retry-scheduling strategy that pairs two disciplines:
• Exponential TTL — An exponential TTL (or more generally an age-based TTL ladder) is a cache-expiry strategy where the TTL assigned to a cached value increases monotonically with the age…
• Exposure bias (ML) — Exposure bias (in the ML-serving sense) is the production hazard where training data and labels are systematically shaped by the production model's decisions…
• Extended DNS Errors — Extended DNS Errors (EDE), defined in RFC 8914, is a supplementary error channel for DNS responses. When a resolver returns a failure (typically SERVFAIL),…
• Extended (R, s, Q) policy — The (R, s, Q) policy is a classical inventory-theory replenishment rule:
• Extensibility in protocol design — Design principle stating that schemas should be structured with foreseeable future change in mind, so that additions and type-evolutions can be accommodated with a new field number…
• Extensible enum — An extensible enum is a string-valued field that carries an allowlist of currently-known values but is explicitly declared at the type level to be open to future additions without…
• External approval ineffectiveness — External approval ineffectiveness names the empirical finding — established by the DORA / Accelerate research program and corroborated by regulator-level studies…
• External credential store — An external credential store is a dedicated database that holds authentication and authorization state — user identifiers, password hashes, role assignments,…
• External engine write to managed table — External engine write to managed table is the architectural shape in which non-vendor compute engines (engines other than the platform vendor's first-party compute) can create,…
• External offset store (CDC) — An external offset store is a CDC-consumer-side durable store that persists the position of the consumer in the source database's change log — binlog / oplog / LSN…
• Externalised authorization — Authorization is externalised when the decision logic (who can do what under which conditions) lives outside application code — in a policy store, a policy engine,…
• FactoryBot object explosion — A test-fixture-library failure mode where declaring an association on a factory implicitly creates a much larger object graph than the test author realised…
• Factuality decoding — Factuality decoding is a category of decode-time interventions that improve the factual accuracy of an LLM without retraining, fine-tuning, or retrieval.…
• Fail-fast principle — Fail-fast is the design principle that when a system cannot satisfy a request within its quality bounds, it should surface the failure as quickly as possible rather than holding…
• Fail-open vs fail-closed — A design choice for what a module does when its input is corrupt, out-of-range, or fails an invariant:
• Fail stale — A failure-mode default where a module, on receiving an input (config file, rule set, feature file, topology snapshot) that fails validation or cannot be read,…
• Failpoint — A failpoint is a labeled, instrumentation-time hook embedded in production source code that can — under test or chaos-driven configuration — pause execution, return an error,…
• Failure-tolerance envelope — The failure-tolerance envelope is the operator-specified bound on the number and kind of concurrent failures a system is engineered to survive without data loss.…
• Fallback cron for self-update safety — Fallback cron for self-update safety is the reliability discipline that when a system X is responsible for its own updates (X applies its own new versions),…
• Fallback experience — A fallback experience is a designated Zalando Experience that the Rendering Engine / domain backend resolves to when the selection-metadata rules produce an ambiguous result…
• False idle detection — False idle detection is the failure mode where a congestion controller (or any rate-adaptive transport component) concludes that its peer / application has stopped sending,…
• False-positive management — False-positive management is the set of practices that keep a continuous-detection system's alert stream signal-dense enough for engineers to trust.…
• False-positive / false-negative asymmetry — False-positive / false-negative asymmetry is the design property of a classification or membership system where the cost of a false positive and the cost of a false negative…
• False-timeout rate — The false-timeout rate is the tunable fraction of requests that will be cut off by a timeout even though the downstream would have eventually succeeded.…
• Fan-in ingestion — Fan-in ingestion is the workload pattern where many independent producers (IoT sensors, application instances, microservices,…
• Fanout and cycle (gossip convergence) — Fanout and cycle are the two numeric knobs that control how fast a gossip protocol spreads a message across a cluster.
• Fast checkpoint via metadata shuffle — A checkpoint / restore implementation in which checkpoint = snapshot the metadata tier; restore = re-point the running workload to a previous metadata snapshot.…
• Fast feedback loops — Fast feedback loops = the architectural property that each proposed code change can be validated (pass/fail, latency, correctness) in seconds rather than minutes or hours.…
• Fast VM boot DX — The developer-experience property that a VM primitive can be treated like a container or a function: started on demand per request / per call / per session,…
• Fat-flow load balancing — Fat flows are network flows characterised by large transfer size and long lifetime — the opposite of the many-short-connections pattern that internet-style traffic typically…
• Fault domain scaling — Fault domain scaling is the challenge that arises when the unit of failure being tested or tolerated grows from a sub-regional fault domain (a rack, a row,…
• Fault-tolerant long-running workflow — A fault-tolerant long-running workflow is any workflow whose expected execution duration is long enough that the probability of encountering some component failure over its run…
• Feature-branch patch management — Feature-branch patch management is the discipline of tracking each internal patch against an upstream OSS project as a named Git feature branch rebased onto each new upstream…
• Feature coverage dashboard — A feature coverage dashboard is the operational instrument that makes the training-serving feature gap visible in production.…
• Feature discoverability — Feature discoverability is the property of a feature store (or adjacent ML data platform) that engineers and modelers can find an existing feature before creating a new one…
• Feature fan-out network bottleneck — Feature fan-out network bottleneck is the failure mode where an online ML serving system that fans out per-candidate feature payloads from a shared feature-fetch tier to multiple…
• Feature-file size limit — A fixed-size cap on the number of rows / features / entries an internally-generated configuration file can contain, enforced at load time by the consumer.…
• Feature flag rollback for validator — Feature flag rollback for validator is the discipline of running a new or stricter validator (typically at admission time, CI,…
• Feature flag — A feature flag is a runtime switch that gates a code path by context, rollout percentage, or targeting rule — enabling release to be decoupled from deploy and cohort-scoped…
• Feature freshness — Feature freshness is the service-level property describing how recently a feature value in an online store reflects the underlying signal that produced it.…
• Feature store freshness — Feature store freshness is the end-to-end latency between a feature-altering event occurring (a click, a purchase, a view) and that event being visible to the feature store's…
• Feature store — A feature store is the class of ML-infrastructure systems that manage and deliver feature data — the numerical/categorical signals a model consumes at both training time…
• Feature taxonomy alignment — Feature taxonomy alignment is the data-engineering activity of ensuring that the same conceptual feature carries the same semantic meaning across source and target domains…
• Federated analytics — Federated analytics is the discipline of computing aggregated, anonymised insights about a population from data that lives only on user devices…
• Federated vs indexed retrieval — Federated retrieval and indexed retrieval are the two high-level architectural choices for building a cross-source agentic retrieval surface.…
• Feed diversification — Feed diversification is the explicit rebalancing of a ranked candidate list to ensure visual, topical, and semantic variety across the served slate…
• Feed-level reranking — Feed-level reranking (a.k.a. slate reranking, feed-composition reranking) is a final reranking stage whose objective is what composition of items to serve,…
• Feedback-control load balancing — Feedback-control load balancing names the class of LB strategies that close a control loop around each backend's observed load: the controller watches per-node utilization,…
• Feedback control loop for rollouts — A feedback control loop for rollouts is the change-management discipline of watching an observable signal as a phased change progresses,…
• Feedback-directed optimization — Feedback-directed optimization (FDO) is the umbrella family of compiler / binary-optimisation techniques where actual runtime execution data is fed back into the compilation /…
• Few-shot prompt template — A few-shot prompt template is a reusable prompt structure that bundles:
• FHIR Implementation Guide — A FHIR Implementation Guide (IG) narrows generic FHIR resource definitions into a use-case-specific conformance contract by adding profiles (constraints on resource shapes),…
• FHIR — FHIR (Fast Healthcare Interoperability Resources) is the HL7 family of standards for exchanging healthcare information electronically.…
• FIDO2 / WebAuthn — FIDO2 is an umbrella standard family from the FIDO Alliance for phishing-resistant public-key authentication. Its two normative components are:
• FIFO cache eviction — First In, First Out (FIFO) is the simplest cache replacement policy. The cache is a queue ordered by insertion time. New items are added at one end;…
• File-based credential store — A file-based credential store is a compliance-substrate pattern where database or service credentials are held in a DBA-provisioned (or operator-provisioned) file on disk rather…
• File-based routing vs bare-HTTP framework taxonomy — File-based routing vs bare-HTTP framework taxonomy is the bifurcation axis that determines how a framework exposes handlers as HTTP endpoints…
• File-class routing — File-class routing is a two-pass document-extraction discipline in which the first pass classifies a document's file class (contract type, schedule type, exhibit type,…
• File-level data skipping — File-level data skipping is the query-planning technique whereby a query engine, given a predicate, eliminates whole data files from the scan set by comparing the predicate against…
• File type spoofing — File type spoofing is the attack pattern where a file presents as one type — via its extension (.jpg) or MIME type (image/jpeg) — while its content bytes are a different type.…
• File vs. Object Semantics — File semantics (the OS filesystem contract applications have been written against for 50 years) and object semantics (the S3-style immutable-blob-with-HTTP-API contract) differ…
• Filesystem as retrieval substrate — Filesystem as retrieval substrate is the agent- architecture choice of storing the knowledge corpus as a filesystem and exposing shell tools (bash, grep, find, cat,…
• Film Grain Synthesis (FGS) — Film Grain Synthesis (FGS) is an AV1 codec tool that removes film grain from the source video before compression, transmits a compact parameter set describing the grain,…
• Filter before A/B test — Filter before A/B test is the experimentation-workflow position in which a cheap pre-filter — typically back-testing, simulation,…
• Filtering-based Internet shutdown — A filtering-based Internet shutdown is a government-directed shutdown implemented at the packet level rather than at the routing layer.…
• Final enum — A final enum is a GraphQL enum value (or an entire enum, by marking all values) annotated with a directive — in Zalando's case @final on ENUMVALUE…
• Fine-grained authorization — Fine-grained authorization means deciding access at the level of individual resources, actions, and runtime context — not just at the level of roles or API endpoints.…
• Fine-Grained Billing — Fine-grained billing is the practice of charging customers at a very small unit of actual consumption (milliseconds of CPU,…
• Fingerprinting vector — A fingerprinting vector is any signal — passive or active — that, when combined with other signals, allows an origin (or a third party) to re-identify a client across sessions…
• FIPS 140-validated cryptographic module — A FIPS 140-validated cryptographic module is a library or HSM whose implementation of cryptographic primitives has been tested by a NIST-certified lab against one of the FIPS 140…
• FIPS cryptographic boundary — FIPS cryptographic boundary is the deployment-topology surface defined by which regions / environments / tenants are required to use only FIPS-140-validated cryptographic modules…
• FIPS mode tri-state — FIPS mode tri-state is a three-valued deployment-altitude config dial exposed by products that support FIPS 140 compliance.…
• Firmware config persistence loss — Firmware config persistence loss is the failure mode in which firmware-level configuration settings — boot order, network-boot interface preferences, secure-boot keys,…
• First AS enforcement — First AS enforcement is a BGP security check specified in RFC 4271 §6.3 where a router verifies that the leftmost AS in a received ASPATH matches the configured peer ASN…
• First-class checkpoint / restore — A sandbox / VM primitive where creating and restoring system-state snapshots is treated as part of the ordinary user workflow, not an escape-hatch-for-emergencies.…
• First-principles theoretical limit — First-principles theoretical-limit reasoning asks: given the physics of the hardware and the structure of the workload, what's the floor on wall-clock time,…
• FIRST test principles — FIRST is Robert C. Martin's acronym (from Clean Code) for the five properties every test — unit, component, integration — should satisfy:
• Fixed-point tree rewriting — Fixed-point tree rewriting is a convergence discipline used by query planners (and compilers) in which a set of rewrite rules is applied to a plan tree repeatedly,…
• Fixed vs variable request cost — Every request to a server has two cost components:
• Flaky test — A flaky test is a test that passes and fails non- deterministically against the same code — producing false positives (fails when the system is working) and false negatives (passes…
• Flamegraph profiling — Flamegraph profiling is the practice of sampling a running process's stack at high frequency and rendering the aggregated stacks as a flamegraph — a horizontal-axis-is-samples,…
• Fleet drain operation — A fleet-ops capability: mark a physical worker as "no more new placements, and evacuate the workloads you currently host somewhere else".…
• Fleet key validation bundle — A fleet key validation bundle is a small, over-the-air-delivered payload that binds a set of fleet public keys to cryptographic signatures from one or more trust authorities,…
• Fleet patching — Fleet patching is the operational capability of a managed-service vendor to deploy a software change (security patch, bug fix,…
• Flexbox as cross-platform layout primitive — Flexbox (CSS Flexible Box Layout) is a web layout model — flex-direction / justify-content / align-items / flex / flex-wrap / ……
• Flink Keyed Stream Union — Flink keyed stream union is the DataStream-API operator-graph shape of DataStream.union(a, b, c, …) → keyBy(k) → process(...): N heterogeneous streams are merged into one typed…
• Flink Snapshot / Savepoint — A Flink savepoint (and its close sibling, the checkpoint) is a consistent, durable snapshot of all operator state in a running Flink job,…
• Flink Stateful Join State Amplification — Flink stateful join state amplification is the pathology of Flink Table API chained joins in which each join operator in a chain maintains its own independent RocksDB-backed state…
• Fluent interface API — A fluent interface (Fowler, 2005) is an API style in which each method returns an object (often this or a related builder) so that calls can be chained into a single expression…
• Flush interval / RPO tradeoff — When an LSM-tree's SSTables and manifest are stored in object storage, the memtable flush interval determines how stale the remote representation can be relative to the in-memory…
• Fly Machine start vs create — Fly.io's Machines API exposes two distinct lifecycle paths that both end with a Machine running user code:
• Follower fetching — Follower fetching is the mechanism by which a consumer reads records from the closest replica of a partition — leader or follower — rather than always reading from the leader.…
• Foreign Iceberg table — A foreign Iceberg table is an Apache Iceberg table whose data and primary metadata catalog live in an external system — AWS Glue, Snowflake Horizon, Hive Metastore,…
• Foreign key constraint — A foreign key constraint is a SQL-schema declaration that a column (or ordered tuple of columns) on a child table must match a corresponding column tuple on a parent table…
• Forged AS_PATH — A forged ASPATH is a BGP ASPATH attribute that has been intentionally manipulated to contain ASNs that do not reflect the actual forwarding path of the route.…
• fork() and execve() — fork() and execve() are the two POSIX system calls that together create and initialise new processes on Unix-family operating systems.…
• Fork-join pool — The JVM's ForkJoinPool is a work-stealing thread pool originally introduced in Java 7 for divide-and-conquer parallelism.…
• Fork-upstream sync — Fork-upstream sync is the operational problem of keeping a privately-modified fork of an open-source project continuously aligned with the upstream project while preserving local…
• Format-aware compression — Format-aware compression is lossless compression where the compressor is given — as an explicit input — the shape of the data it's compressing (rows, columns, enums,…
• Format co-evolution (Iceberg v4 / Delta 5.0) — Format co-evolution is the architectural direction in which two open table formats — Apache Iceberg and Delta Lake — align on a shared core data structure for metadata rather than…
• Format conformance check — A format conformance check is an app-layer validation that untrusted input adheres to its declared format specification — before the app forwards, processes,…
• Forward declaration (C++) — A forward declaration in C++ names an identifier (class / struct / function / enum / alias) without giving its full definition:
• Forward error correction (audio) — Forward Error Correction (FEC) in RTC audio streams means sending redundant copies of (or parity information for) prior audio frames along with current frames,…
• Forward progress — Forward progress is the liveness requirement that a system eventually completes attempted work even when individual participants fail or are partitioned.…
• Forward and reverse edge index — A forward and reverse edge index is a graph-storage technique where every edge is written twice into the link namespace — once keyed by the source node (the forward index) and once…
• Forward security — Forward security (historically forward secrecy in the TLS literature) is the property that compromise of long-term key material at time T does not expose data from sessions…
• Foundational ML platform — A foundational ML platform is a deliberately minimal, reusable core that bolts ML tooling onto a company's production data / compute / orchestration / serving substrate,…
• Four pillars of agent governance — The four pillars of agent governance is Databricks' 2026-05-20 framing for the non-overlapping, stacked architectural surfaces an enterprise needs to govern a non-deterministic,…
• Fragment pruning — Fragment pruning is the design pattern of pushing per-file (per- "fragment", per-"row-group", per-"SSTable") summary metadata up the stack so that a query engine can skip entire…
• Fragmented hardware/software ecosystem — The fragmented hardware/software ecosystem problem is the combined consequence, at the edge-ML layer, of each accelerator vendor shipping its own proprietary ISA,…
• Frame-pointer unwinding — Frame-pointer unwinding is the stack-walking technique that relies on a dedicated CPU register (%rbp on x86-64, x29 on AArch64) storing the caller's frame pointer…
• Fraudulent dispute prediction — Fraudulent dispute prediction is the architectural primitive of predicting whether a payment will result in a fraudulent dispute…
• Fraudulent merchant detection — Fraudulent merchant detection is the architectural primitive of identifying whether a business account (existing or newly-onboarded) on a payments platform poses a fraud risk,…
• Fraudulent website detection — Fraudulent website detection is the architectural primitive of analysing a business's website for fraud red-flags during platform onboarding,…
• Free-space optical communication — Free-space optical communication (FSO) is data transmission through free space — vacuum or atmosphere — using modulated laser beams, rather than through confined media (fibre,…
• Free-trial abuse — Free-trial abuse is the fraud class where a single actor repeatedly cycles through the same product's free trial — by creating new accounts, varying email aliases,…
• Frontend-bound vs backend-bound CPU stall — CPU stalls (cycles where no instruction retires) split into two structural categories at the microarchitecture level:
• Frontier-model batch-training boundary — The frontier-model batch-training boundary is the structural property of the current generation of large language models (2017-2026) that training is exclusively offline batch…
• Frontier-model + local-GPU minion delegation — A hybrid agent-architecture shape where a frontier model (e.g., Claude / GPT-4-class hosted LLM) acts as the orchestrator — planning, routing,…
• FSDP (Fully Sharded Data Parallel) — FSDP (Fully Sharded Data Parallel) is a PyTorch-native distributed-training strategy that shards parameters, gradients, and optimizer state across all data-parallel workers,…
• FUD attack surface — A FUD attack surface is the class of attack where an adversary drains value or trust from a system by spreading fear, uncertainty,…
• Full-dump vs delta vs target (tri-layer CDC schema) — The tri-layer CDC schema decomposes a CDC ingestion job's data into three internal tables, each with a distinct role:
• FULLTEXT index — A FULLTEXT index is MySQL's native text-search index type: a specialised secondary-index shape for CHAR / VARCHAR / TEXT columns that supports word- level search rather than prefix…
• Functional-index collation mismatch — Functional-index collation mismatch is a silent-failure mode in MySQL 8.0.13+ where a functional index is declared on an expression whose result collation differs…
• Functional index (MySQL) — A functional index (also called an expression index) is an index on the result of an expression rather than on a column directly.…
• Fused decompression + matmul — Fused decompression + matmul is the GPU-kernel pattern in which compressed weights are loaded from HBM into shared memory, reconstructed inside SMEM,…
• Fused Int8 ANN search — Fused Int8 ANN search is the GPU-native approximate-nearest-neighbor search primitive used inside Meta's SilverTorch retrieval substrate (Source:…
• Fused multiply-add (FMA) — FMA — fused multiply-add — computes a × b + c as a single hardware instruction with a single rounding step applied only to the final result.…
• Game engine architecture — Game engine architecture is the composition discipline in which an application is built as a small foundation of performance-critical systems (graphics/rendering,…
• Gap identification (top-N cap) — Gap identification with a top-N cap is the agent-output design principle that when a critic or reviewer enumerates missing evidence, uncertainty, or unresolved questions,…
• Gap locking — Gap locking is an InnoDB-specific lock mechanism that locks the space between indexed rows (rather than the rows themselves),…
• Garbage collection (storage) — In immutable / append-only storage systems, garbage collection (GC) is the stage that identifies which blobs, rows, or objects are no longer referenced and marks them safe…
• Gated-community pickup — Gated-community pickup is the ride-sharing UX problem of matching a rider inside a physical access barrier (gate, fob, intercom,…
• Gated schema deployment — Gated schema deployment is PlanetScale's coined term for the deployment-unit model where "all your changes are staged for however long it takes for all of them to be ready"…
• GDPR consent-gated telemetry — GDPR consent-gated telemetry is the practice of gating every telemetry-export call from a customer's browser on their explicit consent signal,…
• Generated column (MySQL) — A generated column is a column whose value is the result of a deterministic expression over other columns in the same row — not a value inserted by the application.…
• Generated hash column — A generated hash column is a MySQL generated column whose expression is a deterministic hash function (MD5, CRC32, SHA2(...…
• Generated knowledge prompting — Generated knowledge prompting is the LLM technique of eliciting intermediate factual or analytical content from the model first,…
• Generative recommendations — Generative recommendations is the class of recommendation- system architectures that use generative AI models (typically LLMs) to produce the content of a recommendation surface…
• Generative retrieval — Generative retrieval is a recommendation / search architecture where the retrieval stage generates the identifier of the next relevant item token-by-token via an autoregressive…
• Geneve tunnel VNI — Generic Network Virtualization Encapsulation (Geneve) is a tunneling protocol commonly used in cloud providers to route a micro-VM's traffic to an isolated tenant network (e.g.,…
• Geographic sharding — Geographic sharding partitions data by a location dimension — usually the user's destination country / region / continent…
• Geospatial sharding — Geospatial sharding is the practice of using geographic location as the shard key for partitioning state across a distributed system.…
• Ghost-node ejection — Ghost-node ejection is the automatic removal of stale node-membership references from a distributed cluster's internal state after a node has left the cluster but before…
• Ghost table migration — A ghost table migration is an online DDL mechanism in which a schema change is applied to a new table (the "ghost" table) that is a copy of the original with the new schema,…
• GIL contention (Python's Global Interpreter Lock) — Python's Global Interpreter Lock (GIL) is the mutex in the CPython reference implementation that ensures only one thread executes Python bytecode at a time.…
• Git-based content workflow — Git-based content workflow replaces a traditional CMS UI with the git repository as the system of record for content. Authors write markdown (or restructured text, MDX,…
• Git delta compression — Git delta compression is how Git makes its pack files compact: rather than storing a full compressed copy of every version of every file,…
• Git pack file — A Git pack file is Git's on-disk compressed bundle of objects. Rather than keep every blob / tree / commit as a loose file under .git/objects/,…
• GitHub Actions script injection — GitHub Actions script injection is the dominant code-execution attack class against CI workflows hosted on GitHub Actions. The vulnerability arises when ${{ github.event.…
• GitOps — GitOps is the operational discipline of treating a Git repository as the single declarative source of truth for system state,…
• Global configuration system — A fleet-wide configuration-delivery channel where a single edit propagates to every server in the fleet within seconds, with no canary, no staged rollout, no per-POP health gating.
• Global routing for databases — Global routing for databases is the architectural idea of routing a client's database connection to the closest edge point-of-presence first,…
• Glossary-constrained translation — Glossary-constrained translation is the discipline of passing a domain glossary (brand terminology, proper nouns, domain terms-of-art) and a placeholder list (i18n template…
• Glossary Dictionary Requirement Translation — Glossary Dictionary Requirement Translation is a three-step methodology for converting ambiguous non-engineering requirements (accountant-voice, legal-voice, product-voice,…
• Glossary term propagation — Glossary term propagation is the automated labeling technique of extending a well-documented column's business glossary term (e.g.…
• Go build tags — Go build tags are file-level compile guards of the form //go:build (Go 1.17+; older syntax // +build ) that tell the Go compiler whether to include a file…
• Go compiler optimization gap — The Go compiler optimization gap is the practical engineering constraint that Go's compiler optimises substantially less aggressively than modern C/C++/Rust compilers (LLVM, GCC).…
• Go plugin dynamic-linking implication — *Importing the stdlib plugin package — even without using it — puts the Go linker into dynamically-linked mode, which disables method dead-code elimination AND forces retention…
• Go runtime memory model (virtual vs resident) — In Go (and in most managed-memory runtimes) the virtual memory the runtime has reserved from the OS and the resident set size (RSS)…
• Goal-oriented orchestrator — A goal-oriented orchestrator is a topology manager that converges a cluster to the state declared by an external source of truth (here: the Vitess topology server) rather than…
• Golden metric selection criteria — Golden metric selection criteria is a governance gate that determines which metrics are important enough to be centrally governed in a semantic layer.…
• Golden record — In Master Data Management, a golden record is "a common, shared, and trusted view on data, also called a golden record, for a particular domain" ().
• Golden signals for RDS — Golden signals for RDS is Zalando's empirical decomposition of RDS Postgres health into twelve metrics across four buckets — CPU, Memory, Disk,…
• Google asset-caching internal heuristics — Google's Web Rendering Service does not obey HTTP Cache-Control headers when deciding whether to re-fetch a page's sub-resources (CSS, JavaScript, images, API responses, fonts).…
• Gossip protocol — A gossip protocol (also called an epidemic protocol — the transmission of messages is analogous to how epidemics spread, or how rumours spread in a crowd) is a family…
• Governance-aware ranking — Governance-aware ranking is the discipline of fusing semantic- similarity scores (retrieval relevance) with governance metadata (table tier, freshness, documentation completeness,…
• Governance travels with resources — Governance travels with resources is the architectural principle — canonicalised in Databricks' 2026-05-20 four-pillars post…
• Governed agent data access — Governed agent data access is a two-axis design surface — access controls (which agent gets what data, on whose behalf, under what consent) + observability (what did the agent…
• Governed SHAP attribution table — A governed SHAP attribution table is a Delta Lake table — registered in Unity Catalog with the same RBAC + lineage + audit posture as any other production table…
• Governed tag — A governed tag is a piece of classification metadata attached to a data-catalog object (catalog, schema, table, column) where the tag itself is a managed primitive: defined once…
• Government-directed Internet shutdown — A government-directed Internet shutdown is a state-ordered suspension of public Internet connectivity, affecting either an entire country, a subset of regions,…
• GPU as hostile peripheral — The security-posture framing that treats a consumer/enterprise GPU as the worst-case hardware peripheral for a multi-tenant platform.…
• GPU kernel utilization — GPU kernel utilization is the percentage of wall-clock time during which the GPU is actively executing compute kernels rather than idle.…
• GPU memory hierarchy — The GPU memory hierarchy is the multi-tier memory layout that any high-performance GPU workload must explicitly reason about.…
• GPU Scale-to-Zero Cold Start — GPU scale-to-zero cold start is the latency penalty paid when inference traffic arrives at a fully-stopped GPU Machine and has to wait for the machine to boot,…
• GPU training failure modes — The specific hardware-failure modes that dominate interruptions in large GPU training fleets. Meta's 2024-06-12 post enumerates three as the most frequent modes observed at their…
• Graceful degradation — Graceful degradation is the design property that when a subsystem or dependency fails, the service as a whole continues to operate in a reduced-but-useful mode,…
• Graceful saturation vs. congestive collapse — Graceful saturation is the substrate property where a system at 100% resource utilisation plateaus — throughput stops growing, latency climbs predictably,…
• Gradient-free black-box optimisation — Gradient-free black-box optimisation is the class of optimisation methods that treat the objective f(θ) as an opaque function — you can call f(θ) and get a value,…
• Grain intensity scaling function — A grain intensity scaling function is a piecewise-linear function from pixel value → grain intensity transmitted alongside AR grain coefficients in the AV1 Film Grain Synthesis…
• Grain misalignment — Grain misalignment is the data-engineering antipattern of running a single pipeline at the finest grain that any consumer needs,…
• Granular usage attribution — Granular usage attribution is the observability primitive of tracking who reads what data, at what granularity finer than the table or bucket level…
• Granularity-aligned bucket — A granularity-aligned bucket is a fixed-size time bucket whose boundaries are aligned to a query's declared granularity (1 minute, 5 minutes, 1 hour).…
• Graph traversal fanout — Graph traversal fanout is the number of edges a traversal expands across at each hop. For a 1-hop traversal from a source node, fanout is the source's degree.…
• Graphics API abstraction layer — A graphics API abstraction layer is the interface a rendering codebase presents to the rest of its application, separate from the underlying graphics API (WebGL, WebGPU, Metal,…
• GraphQL error extensions — GraphQL error extensions are an open-ended metadata channel on each error object in the response.errors array. The base envelope GraphQL guarantees is fixed:
• GraphQL error propagation — GraphQL error propagation is the runtime semantic defined by the GraphQL specification: when a resolver throws an error on a non-nullable field,…
• GraphQL persisted queries — GraphQL persisted queries is the technique of registering each GraphQL query that an application might issue ahead of time (typically at UI build time) and replacing the query text…
• GraphQL query directive — A GraphQL query directive (spec term: executable directive) is a directive whose declaration permits it on one or more ExecutableDefinition locations…
• GraphQL schema directive — A GraphQL schema directive is a directive whose declaration permits it on one or more TypeSystem locations — the parts of the schema that describe types rather than executable…
• GraphQL schema usage observability — GraphQL schema usage observability is the property — only cheaply available once the production query set is closed and enumerable — that for every field in the schema,…
• Greenfield developer app for hybrid engineers — A greenfield developer app (in the brownfield-RN context) is a parallel, standalone React Native application built and maintained alongside the main legacy app.…
• Grep loop — The grep loop is an agent failure mode where a documentation set that exceeds the agent's context window forces the agent into iterative grep-style keyword search across the corpus…
• Grey failure — Grey failure names a component that is not fully broken but not fully healthy — partially, intermittently, or sub-specification degraded.…
• Group-based access control (GBAC) — Group-Based Access Control (GBAC) is an authorization model where permissions are granted to groups (not individual users) and group membership is provided by a centralized…
• GSI cost anti-pattern at petabyte scale — A DynamoDB Global Secondary Index (GSI) stores a projection of the base table keyed on a different attribute set, and is billed separately for storage + write capacity.…
• GTID position — A GTID (Global Transaction Identifier) in MySQL is a unique label for every committed transaction, of the form :.…
• H2 vs real database testing — A recurring testing-infrastructure choice: run JVM tests against an embedded, in-memory SQL emulator (classically H2, also HSQLDB / Derby) configured in MySQL/PostgreSQL…
• HA / CDC coupling — HA / CDC coupling is an operational coupling between a database cluster's high-availability actions (primary promotion, replica replacement,…
• Hairpin route leak (Type 1) — A Type 1 hairpin route leak (per RFC 7908) is the most common and simplest route-leak shape: a multi-homed customer takes routes from one of its providers and re-advertises them…
• Hard-drive physics (capacity vs. seek-time) — Hard drives are mechanical devices (spinning platters + moving arm + flying head). This constrains them in a way that grows more severe every generation: capacity scales fast,…
• Hardcoded literal address antipattern — The hardcoded literal address antipattern is the practice of writing literal IP addresses (or raw endpoint identifiers) into configuration where a name — DNS, a service registry,…
• Hardware-accelerated video codec API — A hardware-accelerated video codec API is a standardised interface inside a media toolchain that lets multiple vendors' fixed-function video encoder/decoder hardware be driven…
• Hardware-aware model architecture — Hardware-aware model architecture is a model-design discipline where the model's structural choices are deliberately aligned with the underlying hardware's capabilities…
• Hardware-isolated micro-VM on Kubernetes — Compose Kubernetes (scheduling, networking, declarative API, lifecycle) with Firecracker (hardware-virtualisation boundary per workload) so that each scheduled "pod" / workload…
• Hardware offload — Hardware offload is the design pattern of moving work that previously ran on general-purpose CPUs under a general-purpose OS/hypervisor onto dedicated hardware pipelines…
• Hardware Proprietary Knowledge Injection — Hardware proprietary knowledge injection is the architectural mechanism for making code-generation models productive against hardware whose documentation is not in public…
• Hardware reliability at scale — Hardware reliability at scale is the property of a large hardware fleet (thousands+ of GPUs / servers) that makes a single long-running job — a training run, a batch query,…
• Hardware security module (HSM) — A hardware security module (HSM) is tamper-resistant cryptographic hardware whose primary job is to generate, store, and use cryptographic keys inside a boundary the host operator…
• Hardware/software co-design — Hardware/software co-design is the practice of shaping hardware requirements and software design together, as one iterative loop, rather than in sequence.…
• Harvest-now, decrypt-later (HNDL) — Harvest-now, decrypt-later (also store-now-decrypt-later / steal-now-decrypt-later) is the threat model in which an adversary captures encrypted traffic today and stores it…
• Hash-collision embedding tradeoff — The hash-collision embedding tradeoff is the core tension in sizing embedding tables for sparse categorical features in recommendation models:
• Hash ring — The hash ring is the topology used by consistent hashing to lay out nodes and keys: the output of the hash function is treated as a fixed circular address space in which…
• Hash sharding — Hash sharding routes each row to a shard by hash(shardkey) — a deterministic hash function maps every shard-key value into a fixed output range,…
• HBM vs. SMEM (GPU memory hierarchy) — Modern NVIDIA GPUs have two relevant memory tiers for inference kernels:
• HDD sequential I/O optimization — HDD sequential I/O optimization is the design stance of laying data out on disk such that the access pattern is linear — contiguous byte runs…
• HDR10+ — HDR10+ is an open, royalty-free high-dynamic-range video format that extends the static-metadata HDR10 baseline with dynamic per-scene metadata describing peak luminance and colour…
• Head-of-line blocking — Head-of-line (HOL) blocking occurs when a slow item at the front of an ordered queue or stream prevents subsequent items from being processed,…
• Head-of-Line Buffering (in streaming pipelines) — Intermediate layers — reverse proxies, compression middleware, CDN edges, buffer-mode stream transforms — typically default to waiting until a size or time threshold is reached…
• Header-based mock switching — Header-based mock switching is the technique of using an L7 routing layer (an ingress proxy, a service mesh, or a sidecar) to inspect an inbound request header and route the call…
• Headless API invocation — Headless API invocation is the property that a tool can be driven without a graphical user interface — typically via a programmatic SDK (Java / Python / etc.) or a command-line…
• Headless BI semantic layer — A headless BI semantic layer is a single, governed, tool-agnostic layer in the data platform where business metrics and KPIs are defined once,…
• Headless CMS — A headless CMS is a content management system that separates authoring from delivery. It provides a structured authoring UI and a content data model,…
• Headless IDE inspection — Headless IDE inspection is the technique of taking an IDE tool that was designed to run interactively — a refactor, an inspection,…
• Health-mediated deployment — A deployment methodology where a change — code or configuration — is rolled out progressively across stages gated by real-time health signals,…
• Healthcare data interoperability — The healthcare data interoperability problem is the structural challenge of building a unified, governed, queryable view of patient data when the source data lives in dozens…
• Heap-dump lock introspection — Heap-dump lock introspection is the diagnostic technique of reading the on-heap state of a lock object to determine its current owner and waiter queue…
• Heap-organized table — A heap-organized table (also heap table, unordered heap) stores rows in a pile of unordered pages with no built-in physical sort order.…
• Heartbeat-based ownership — Heartbeat-based ownership represents the owner of a mutable resource (an IP address, a partition, a capacity slot) as a list of non-overlapping (owner, tstart, tend) time ranges,…
• Heartbeat counter (gossip) — The heartbeat counter is the local liveness signal that a gossip node attaches to every state exchange. It is the data-plane half of gossip-based failure detection: stale counters…
• Heat management (storage) — Heat in a multi-tenant storage system is the number of requests hitting a given drive per unit time. Heat management is the ongoing placement and steering problem of keeping…
• Hermetic build — A hermetic build (or test) declares every input it depends on — source files, tool binaries, environment variables, OS libraries…
• Heterogeneous AI Accelerator Fleet — A heterogeneous AI accelerator fleet is a production AI infrastructure spanning multiple vendors (NVIDIA, AMD, custom proprietary silicon,…
• Heterogeneous code migration — Heterogeneous code migration is the failure mode where a body of code that needs to be migrated from one platform to another is so non-uniform across instances that no fixed set…
• Heterogeneous data formats — Heterogeneous data formats are the real-world data mix agents must handle: CSV + JSON + markdown + plain text + logs + binary — structured, semi-structured,…
• Heterogeneous fleet config skew — Heterogeneous fleet config skew is the failure mode where different hosts in the same fleet accumulate different configuration defaults over time — different package versions,…
• Heuristic retrieval — Heuristic retrieval is a stage-1 retrieval primitive that uses domain-specific, non-ML rules — ownership metadata, graph traversal, time windows, string matching,…
• Hexagonal architecture — Hexagonal architecture (Alistair Cockburn, 2005), also known as ports-and-adapters, organizes a codebase into concentric layers where the innermost domain holds business rules…
• Hidden agent directive — A hidden agent directive is a short textual instruction embedded in an HTML page that is visible to LLM crawlers / parsers but invisible to human readers.…
• Hierarchical batch sampling for contrastive loss — Hierarchical batch sampling is the technique of constructing training batches deliberately from a taxonomy structure so that a catalog-tree contrastive loss always has positive…
• High-cardinality aggregation overload — A high-cardinality aggregation overload is the failure mode where an aggregation query (terms, cardinality, significantterms) is executed against a field whose distinct- value…
• High-cardinality attribute indexing over object storage — A class of indexing technique that makes unique-value lookups fast over datasets whose bulk storage lives on object storage (S3, GCS,…
• Highest Priority Content (HPC) — Highest Priority Content (HPC) is GitHub's internal client-side rendering metric that measures when the primary content on the page is first rendered…
• Historical pickup heatmap — A historical pickup heatmap is a spatial aggregation of where, in a given area, riders and drivers have actually successfully met in past trips.…
• HNSW index — An HNSW (Hierarchical Navigable Small World) index is a multi-layer proximity-graph ANN index with state-of-the-art recall / latency for in-memory vector search.…
• Holistic failure detection via replicas — Holistic failure detection (in the Orchestrator sense) is a primary-outage detection scheme that triangulates the failure observer's own reachability to the primary…
• Horizon operator — The Horizon operator is a placeholder operator in the VTGate query planner's operator tree that packages together the entire "after-the-FROM-clause" portion of a SQL query…
• Horizontal sharding — Horizontal sharding splits a single logical table (or group of related tables) so that its rows live across multiple physical database instances.…
• Horizontal vs vertical scalability for feature engineering — Horizontal scalability adds more worker nodes to spread load; each worker handles a share of the data. Canonical technologies: Spark, distributed SQL engines, MapReduce.
• Host-consistency sliding upgrade — Host-consistency sliding upgrade is the rollout discipline of partitioning the host software stack into two layers and treating them with opposite policies:
• Hosted vs local MCP server — Hosted vs local MCP server is the deliberate architectural-choice axis between running an MCP server as:
• Hot cluster reuse — Hot cluster reuse is the practice of keeping compute clusters (CPU or GPU) active and pre-provisioned between workflow runs,…
• Hot-cold code splitting — Hot-cold code splitting is a compiler / binary-optimiser transformation that separates frequently-executed and rarely-executed parts of a function (or of the whole binary) into…
• Hot-cold tier compression codec split — A storage-tiering discipline where different compression codecs are applied per tier according to the query-access profile: recent/hot data uses a lightweight codec (decode-fast,…
• Hot key — A hot key is a single key whose request rate is disproportionately higher than the rest of the keyspace. Under any scheme that maps one key to one node (static-sharding /…
• Hot path — The hot path is the code that runs on every (or near-every) request, especially at high request rates. It's the call chain whose per-invocation cost is multiplied by the request…
• Hot-reloadable configuration — Hot-reloadable configuration is a process's ability to pick up a new configuration value without a restart. The test: you change a setting (retry count, connection pool size,…
• Hot-reloading devloop — Hot-reloading devloop is the developer-experience property where code changes become visible in a running application within sub-seconds of saving the file, with no compile step,…
• Hot-row problem — The hot-row problem is the data-shape pattern where one row in a table receives a disproportionate fraction of the workload's writes (or reads)…
• Hot row protection — Hot row protection is a Vitess query-serving safety feature that limits the concurrent write amplification against a single hot row before the contention reaches the underlying…
• Hot shard write frontier — A hot shard write frontier is the specific failure mode where all new inserts into a horizontally-sharded table target the same shard because the shard key is monotonic…
• Hotspot — A hotspot is a performance-degraded node in a distributed system that takes a disproportionate share of data storage or request traffic.…
• HOV Lane (High-Occupancy Vehicle Lane) — An HOV lane (high-occupancy vehicle lane; colloquially carpool lane) is a lane on a road network reserved for vehicles carrying multiple passengers (thresholds vary…
• HPKE (Hybrid Public Key Encryption) — HPKE (RFC 9180) is a standardised primitive for hybrid public-key encryption: the sender uses an asymmetric key encapsulation mechanism (KEM) to establish a shared secret…
• HTAP (Hybrid Transactional / Analytical Processing) — HTAP is a database category promising to serve both OLTP and OLAP workloads from a single system — collapsing the traditional two-system split (transactional database + separate…
• HTTP/3 probing gap — The HTTP/3 probing gap is the observability failure class that appears when an org rolls out HTTP/3 on its edge: existing client-side black-box probers…
• HTTP/3 — HTTP/3 is the HTTP-over-QUIC version of HTTP — standardised in RFC 9114 (2022). Unlike HTTP/1.1 and HTTP/2 (which run over TCP), HTTP/3 runs over QUIC,…
• HTTP 402 Payment Required — HTTP 402 Payment Required is an HTTP client-error status code defined in the HTTP/1.1 standard as "reserved for future use"…
• HTTP Message Signatures (RFC 9421) — HTTP Message Signatures (published as RFC 9421 in 2024) is an IETF standard for attaching one or more cryptographic signatures over a canonical representation of an HTTP request…
• Hub-and-spoke broadcast architecture — A hub-and-spoke broadcast architecture terminates N venue-side contribution feeds ("spokes") at a single central facility ("hub") that handles signal ingest + inspection +…
• Hub-and-Spoke Governance — Hub-and-spoke governance is the posture where a single central catalog owns metadata + access-control policy + identity for the whole mesh ("hub"),…
• Huffman coding — Huffman coding (Wikipedia) is a variable-length prefix code that assigns short bit codes to common symbols and long bit codes to rare symbols.…
• Huge-page allocator (HPA) — A huge-page allocator is the subsystem of a userspace memory allocator responsible for serving allocations backed by huge pages (typically 2 MiB or 1 GiB on x86-64;…
• Human–LLM evaluation alignment — Human–LLM evaluation alignment is the measured agreement between an LLM judge's verdicts and a human rater's verdicts on the same inputs using the same rubric.…
• Hybrid back-testing with ML counterfactual — Hybrid back-testing is the pattern of replaying historical system state through alternative code paths and using ML models to predict the outcomes that would have occurred under…
• Hybrid clean/noisy training corpus — A hybrid clean/noisy training corpus is a pre-training dataset deliberately composed of two tiers:
• Hybrid cloud media ingest — Hybrid cloud media ingest is the infrastructure shape of placing high-bandwidth ingest capacity physically close to production sites while using a large cloud provider…
• Hybrid key encapsulation — Hybrid key encapsulation is a key-agreement construction where two independent KEMs run in parallel and the resulting session key is derived from both of their shared secrets…
• Hybrid ML platform architecture — Hybrid ML platform architecture is the shape in which a single ML platform splits its compute tier from its serving tier onto different substrates chosen to match each workload's…
• Hybrid multi-tenant architecture — Hybrid multi-tenant architecture names the class of SaaS deployment shapes that mix isolated and shared infrastructure at different layers of the stack rather than committing…
• Hybrid retrieval (BM25 + dense vectors) — Hybrid retrieval is the pattern of combining a lexical index (BM25 / keyword) with a dense vector index (semantic embeddings) in the same retrieval pipeline,…
• Hybrid Search (vector similarity + metadata filter) — Hybrid search (in the vector-DB sense) is the retrieval primitive that combines a vector similarity search with a structured filter over the vectors' metadata fields,…
• Hybrid tower inference split — Hybrid tower inference split is a serving-time architectural split applied to the user (or query) tower of a two-tower retrieval model: the expensive,…
• Hybrid Vector Tiering (Cold S3 ↔ Hot OpenSearch) — Hybrid vector tiering is the storage-and-query pattern that recognises vector workloads have bimodal access profiles — a large slow-growing archival set where storage cost…
• Hybrid vs replacement PQC deployment — When integrating post-quantum cryptography into a protocol or system, there are two deployment paths:
• HyDE (Hypothetical Document Embedding) — HyDE (Hypothetical Document Embedding) is a retrieval technique where, instead of embedding the user's question and searching for documents similar to the question,…
• Hydration mismatch — A hydration mismatch is the failure mode where a React (or similar SSR-then-CSR framework) app produces different markup on the server than on the client for the same component,…
• Hyperscale compute workload — A hyperscale compute workload is the shape of application compute that runs in large-scale internet-company datacenter cloud deployments — the workload population served by Meta,…
• Hypothetical-index evaluation — Hypothetical-index evaluation is the practice of costing a proposed database index without actually creating it, by registering a hypothetical index structure with the query…
• Iceberg catalog REST sync — Iceberg catalog REST sync is the mechanism by which a producer (streaming broker, batch writer, CDC connector) registers and maintains a table in an external Apache Iceberg REST…
• Iceberg file-based catalog — An Iceberg file-based catalog is a catalog-integration shape in which a Apache Iceberg-writing producer publishes its table's current snapshot pointer as a metadata file directly…
• Iceberg snapshot expiry — Iceberg snapshot expiry is the metadata-garbage-collection loop that prunes old Iceberg snapshots (and the manifest + data files they uniquely reference) as the table accumulates…
• Iceberg topic mode — Iceberg topic mode (Redpanda topic-level configuration redpanda.iceberg.mode) selects how the broker projects a Kafka record into an Iceberg row when Iceberg Topics is enabled…
• Iceberg topic — An Iceberg topic is a streaming-broker construct — introduced by Redpanda in 2024 — in which a single logical entity is both a Kafka-protocol topic and an Apache Iceberg table…
• Idempotency token — An idempotency token is a per-operation identifier attached to a mutation so that a retried or hedged write is logically indistinguishable from the original…
• Idempotent build action — A build action is idempotent if, given the same declared inputs, it always produces the same output. It's a necessary condition for sound caching: if running the same action twice…
• Idempotent job design — An idempotent job is one that can be run multiple times for the same input and produces the same end-state as running it once.…
• Idempotent migration — An idempotent migration is a database schema change script that produces the same end state regardless of how many times it is executed against the same database.…
• Idempotent operations — An idempotent operation is one whose effect is the same whether executed once or many times with the same inputs. Idempotency is the enabling property for safe retries: without it…
• Idempotent thread-safe order-agnostic scan-step — An idempotent thread-safe order-agnostic scan-step is the load-bearing worker contract that batch-processing frameworks require of their per-batch workers,…
• Identity decoupling (user ID vs profile ID) — Separating the internal user entity (the complete record: legal name, email, phone, account state) from one or more public-facing profiles (context-scoped subsets of that record).…
• Identity graph — An identity graph captures relationships between users and their associated entities (devices, emails, phone numbers, payment methods, IP addresses) in a graph structure.…
• Identity-hiding handshake — An identity-hiding handshake is a cryptographic handshake whose wire form does not disclose the long-term identity of either endpoint to a passive observer.…
• Identity vs behavior proof — Identity proof and behavior proof are two distinct answers to the question "should I trust this request?". Cloudflare's 2026-04-21 post argues that most web-protection decisions…
• if-let lock-scope bug — In Rust, an if let expression that acquires a lock in its scrutinee holds that lock across both the if arm and the else arm…
• IID disruption from request-sorted data — IID disruption is the failure mode that appears when training-dataset rows are no longer independent and identically distributed across a batch,…
• Immediate encoding limit — Immediate encoding limit is the architectural constraint on how wide a value can be encoded directly inside a single instruction.…
• Immutable aggregation window — An immutable aggregation window is a time range closed for further writes, inside which an event-log aggregation can run without coordination and still converge to the same value…
• Immutable index state — Immutable index state is a state-management discipline for search engines (and any similarly-shaped service) in which the current state is held in an immutable object,…
• Immutable Object Storage — Immutable object storage is a model in which the stored unit — the object — cannot be partially modified after it is written.…
• Immutable partition — An immutable partition is a Cassandra (or wide-column-store) partition that has provably stopped receiving new writes — typically because the partition's identifying time range has…
• Immutable segment file — An immutable segment file is an on-disk unit of index storage that is written once, never modified, and eventually replaced (via merging) or deleted (via garbage collection)…
• Implicit prompt caching — Implicit prompt caching is the design choice that an LLM-serving platform performs KV-cache reuse across requests automatically, without requiring the caller to opt in,…
• Implicit query location rewrite — Implicit query location rewrite is the search-pipeline technique of rewriting the location portion of a user's search request (typically the geobox…
• In-band signaling handshake — In-band signaling is the exchange of configuration / capability / policy information between a client and a server via the same RPC channel they use for request traffic…
• In-batch negative false-negative — In contrastive two-tower retrieval training, the in-batch negative sampling trick uses other candidates in the same training batch as negative examples for each anchor…
• In-cluster streaming SQL — In-cluster streaming SQL is the architectural property where the analytical query engine is deployed inside the same cluster as the streaming broker and storage substrate,…
• In-Context Reinforcement Learning — In-context reinforcement learning is a learning mechanism in which an agent's capability compounds over time through writes to a persistent retrieval store consulted at inference…
• In-kernel filtering — In-kernel filtering is the architectural move of evaluating per-event match/drop decisions inside the kernel (via ebpf), before events are handed off to user-space consumers.…
• In-loop quality metrics — In-loop quality metrics are reference visual-quality metrics (e.g. PSNR, SSIM, VMAF) computed during transcoding — inside the same command producing the encoded output…
• In-memory coalescing by Kafka key — In-memory coalescing by Kafka key is the technique of merging multiple Kafka messages that share the same key into a single aggregate inside a consumer's batch buffer before…
• In-memory schema metadata graph — An in-memory schema metadata graph is a runtime data structure that materialises a graph database's schema as a small graph in process memory,…
• In-memory tenant state — In-memory tenant state names the service property where tenant-specific data is loaded into RAM at process startup and served from memory,…
• In-place vs new-DC database upgrade — When upgrading a distributed datastore to a new major version, two architectural shapes are available:
• In-source CDC checkpointing — In-source CDC checkpointing is the offset-durability class where a CDC consumer persists its progress position inside the source database itself…
• In-sync replica set (ISR) — For each Kafka partition, the in-sync replica set (ISR) is the subset of the partition's replicas that are caught up with the leader's log.…
• Incident mitigation lifecycle — An incident mitigation lifecycle is the end-to-end process by which an emergency protective control — typically a rate limit, block rule, feature flag,…
• Incident playbook — An incident playbook is a written, structured, pre-approved emergency procedure that an incident responder can execute to put a system into a known degraded state during an outage…
• Incomplete request — An incomplete request is a request for which the elector (during a leadership change) cannot determine whether the old leader had made it durable.…
• Incremental backup on commit — Incremental backup on commit is a backup discipline in which every commit of a system's durable state triggers a backup of only the files that changed since the last commit,…
• Incremental delivery (over big-bang rewrites) — Incremental delivery is the posture of shipping architectural change as a sequence of small, observable, reversible steps — each of which delivers some customer-visible value…
• Incremental indexing — Incremental indexing is the discipline of processing just the changes to a codebase when updating a code index — target complexity O(changes) instead of O(repository).…
• Incremental Static Regeneration — Incremental Static Regeneration (ISR) is a caching strategy — popularised by Next.js — in which a page is:
• Incremental vector index — An incremental vector index is an ANN index that absorbs point mutations — inserts, updates, deletes — continuously, without periodic full rebuilds,…
• Incremental View Maintenance (IVM) — Incremental view maintenance (IVM) is the database technique of keeping a materialized view up to date by applying only the deltas induced by changes to its inputs,…
• Index as Model — Index as Model is the architectural paradigm — coined by Meta in the 2026-05-26 SilverTorch post (Source:…
• Index-candidate filtering — Index-candidate filtering is the practice of using workload telemetry to decide which query patterns are candidates for indexing at all,…
• Index cardinality — Cardinality is the count of distinct values in an indexed column — computed as COUNT(DISTINCT col). It is a raw number (e.g. 3,009 distinct first names), not a ratio.…
• Index obfuscation — Index obfuscation is the situation where a query filters on an indexed column but wraps the column in an expression (function call, arithmetic,…
• Index selectivity — Selectivity is the fraction of a table's rows that a query predicate against an indexed column is expected to match — expressed as COUNT(DISTINCT col) / COUNT(),…
• Index-usage time series — An index-usage time series is a per-query-pattern record, over time, of which indexes the database used to execute the pattern. The axes:
• Index write cost — Every secondary index is a tax on writes. Adding an index speeds up reads that can use it and slows down every INSERT and UPDATE that touches an indexed column.…
• Inference compute–storage–network locality — For transaction-shaped inference workloads, the load-bearing architectural axis is the combination of GPU compute + instance RAM + fast object storage for model parameters/datasets…
• Inference Pipeline Model — An inference pipeline model is a model-serving composition where a single endpoint / service exposes a chain of containers (or processing stages) that run in-process on the same…
• Inference trilemma (recsys) — The inference trilemma in production recsys serving is the three-way tension between:
• Inference vs training workload shape — Inference and training are fundamentally different workload shapes, and the infrastructure design that fits each is correspondingly different.…
• Information flow control (IFC) — Information flow control (IFC) is the classical security / privacy primitive that controls how data is processed and transferred at runtime — not just who can read it at rest.…
• Init-container IP-gossip pre-migration — A discipline for upgrading a gossip-based distributed datastore on Kubernetes where pods get new IPs on restart: sequence the IP change and the version change into two distinct…
• Injection bandwidth (AI cluster) — Injection bandwidth (in the context of an AI training cluster) is the per-accelerator network bandwidth at which a single GPU/accelerator can inject data into the fabric.…
• Inline LLM content guardrail — An inline LLM content guardrail is a runtime control surface that scans the content of every LLM request and response on the inference critical path (not async post-processing)…
• Inner container VM — A Linux container slid between user code and the VM's kernel, separating:
• Inner sourcing — Inner sourcing (or innersource) is the application of open-source-style contribution practices inside a single company. Teams that do not own a platform contribute changes to it…
• InnoDB buffer pool — The InnoDB buffer pool is an in-memory cache of InnoDB pages (default 16 KB each) sitting between MySQL query execution and the on-disk B+tree pages.…
• InnoDB index_init hook — MySQL's storage-engine abstraction layer — the handler API — lets each storage engine plug into a fixed set of callbacks the server invokes during query execution.…
• InnoDB internal operations table — An internal operations table is a PlanetScale-fork-specific MySQL / InnoDB primitive that marks specific tables as foreign-key-exempt: InnoDB skips all FK validation and cascading…
• InnoDB page fill factor — The page fill factor is the fraction of a B+tree leaf page InnoDB fills with row data before it splits the page and allocates a new one.…
• InnoDB parent-table rename pinning — InnoDB parent-table rename pinning is the PlanetScale-MySQL-fork behaviour — enabled by setting the server variable renametablepreserveforeignkey=1…
• InnoDB silent cascade in binlog — InnoDB silent cascade in binlog is the architectural property — specific to MySQL + InnoDB — that when a parent-side DELETE or UPDATE fires foreign-key cascading actions (ON DELETE…
• Input image selection tradeoff — When a vision-language model (VLM) pipeline ingests images to extract attributes, the specific image(s) chosen from the available set is a first-order lever on output quality…
• Inquiry Type classifier — An Inquiry Type classifier is a small fine-tuned model that decides whether a user question is in-scope for an LLM product's domain,…
• Inside-out orchestration — A VM-platform architectural pattern in which the majority of orchestration and management services run inside the VM itself (in the VM's root namespace) rather than on the physical…
• Instance-type fallback (prioritised node affinity) — Instance-type fallback is the scheduling pattern where a workload declares a prioritised list of acceptable compute instance types and the scheduler silently places the pod…
• MySQL INSTANT DDL — MySQL ALGORITHM=INSTANT is a ALTER TABLE execution strategy introduced in MySQL 8.0 (with significantly expanded coverage in 8.0.29) that applies a schema change by updating only…
• Instant-deploy eligibility — Instant-deploy eligibility is the three-part composition rule PlanetScale applies at deploy-request submission time to decide whether a schema change can be executed via MySQL's…
• Instantaneous power loss — Instantaneous power loss is a zero-notice disaster scenario in which a data center region loses all power supply without any advance warning — no time for graceful shutdown,…
• Institutional memory — The organisational property that an engineering org retains the lessons learned from its past incidents, design decisions,…
• Instruction-cache locality — Instruction-cache locality is the property that a program's hot execution path fits in the CPU's instruction cache + iTLB and is laid out contiguously enough that sequential…
• Instruction dispatch cost — Instruction dispatch cost is the per-instruction overhead a bytecode or VM interpreter pays to fetch the next opcode, decide which handler to run,…
• Instrumentation engineering cost — The instrumentation engineering cost is the human engineering effort required to add measurement to a system — per-surface, per-service, per-handler, per-subsystem.…
• Instrumentation-storage decoupling — Instrumentation-storage decoupling is the architectural property of using a stable, portable protocol (OpenTelemetry / OTLP) as the boundary between the agent / service emitting…
• Instrumented vs sampling profile — There are two structural shapes for collecting the profile data that feeds PGO, LLVM BOLT, and similar feedback-directed optimisation tools:
• Insurgent cloud constraints — The set of structural disadvantages that a non-hyperscaler cloud (an "insurgent cloud" — Fly.io, Render, Railway, Vultr, Hetzner in one frame;…
• Integer overflow — Integer overflow is the condition in which the result of an arithmetic operation on fixed-width integer types falls outside the representable range of that type.…
• Integration tests against real database — Integration tests against real database is the testing discipline in which a test case runs against a live instance of the actual database technology that production uses…
• Intelligent Document Sampling — Intelligent Document Sampling is a cost-reduction strategy for LLM-driven document classification: instead of running the model over every page of a long document,…
• Interaction to Next Paint (INP) — Interaction to Next Paint (INP) is a Core Web Vital that measures the latency between a user interaction (click, tap, key press) and the browser painting the next frame…
• Interdependent systems — Interdependent systems are architectures in which a change in one subsystem can surface as a bug in an entirely different subsystem, because the two share data, state,…
• Interlanguage null safety — Interlanguage null safety is the set of mechanisms — annotations, static analysers, and runtime checks — that keep null-related invariants intact when control crosses a language…
• Interleaving testing — Interleaving testing is an online-evaluation technique for ranking experiments. Instead of splitting users between two rankings (A/B testing),…
• Intermediate Fusion (Multimodal) — Intermediate fusion is the multimodal-ML strategy of encoding each modality separately through a modality-specific encoder,…
• Intermediate representation bottleneck — A multi-stage pipeline that stages its data through a fixed intermediate representation can end up with that intermediate being strictly less expressive than either the upstream…
• Intermittent failure signal confusion — An incident pattern where the shape of the failure signal (intermittent, oscillating, spiky) mimics an external attack more than an internal bug…
• Internal fork divergence — Internal fork divergence — colloquially "the forking trap" — is the monotonic process by which a company's internal fork of a large open-source project drifts apart from upstream…
• Internally-generated untrusted input — A trust-boundary confusion where an engineering org treats data it generated itself (config files, feature files, topology snapshots,…
• Internet-scale interoperability — Internet-scale interoperability is the property a cryptographic protocol (or any networking protocol) must satisfy to be deployed across the heterogeneous, multi-vendor,…
• Intersecting quorums — Intersecting quorums is the principle that the correctness of a single-leader consensus system depends on every write-quorum intersecting every read/election-quorum on at least one…
• Intertoken latency (ITL / time-per-token) — Intertoken latency (ITL), also called time-per-output-token (TPOT), is the LLM-serving latency metric measuring the delay between successive output tokens during streamed…
• Invalid route observability metric — An invalid route observability metric is a per-label-set counter emitted by a validator for every rejection, exposed before the validator's rejections actually block writes,…
• Invalidation-based cache — An invalidation-based cache keeps cached values until notified they might be stale, at which point the cached entry is evicted and (on next read) re-fetched from source.…
• Inverted-index deduplication — A scaling technique for deduplicating AI-generated findings at fleet scale by using deterministic code to build inverted indexes over structured data,…
• Investigation-phase progression — Investigation-phase progression is the architectural discipline of modelling a long-running agent loop as a sequence of named phases (e.g.…
• I/O-latency-sensitive workload — A workload whose end-to-end performance is bounded by the per-I/O round-trip latency of the storage substrate rather than by aggregate IOPS throughput or sequential bandwidth.…
• IO wait — %iowait on Linux is the fraction of CPU time reported as idle while at least one CPU-local task is blocked on disk I/O. It appears as the wa column in vmstat and top,…
• IOPS (Input/Output Operations Per Second) — IOPS — Input/Output Operations Per Second — is the rate at which a storage device performs read or write operations against its backing media.…
• IOPS throttle on network-attached storage — Cloud providers that expose network-attached block storage (Amazon EBS, Google Persistent Disk, etc.) cap the number of I/O operations per second a volume can accept,…
• IP address fragmentation — IP address fragmentation is the phenomenon where a cluster's available IP address space is distributed across many small subnets such that no single subnet has enough contiguous…
• IP attribution — IP attribution is the problem of mapping a packet's or flow's IP address (and the time at which that IP was in use) to the workload identity that owned it at that moment.…
• IP rotation as evasion — IP rotation in the evasion sense is the practice of spreading automated-client requests across many source IPs — residential proxies, VPN exit nodes, cloud-egress pools,…
• IP warming — IP warming is the email-marketing practice of gradually increasing daily send volume from a new sending IP over a period of weeks so that receiving ISPs (Gmail, Outlook, Yahoo,…
• IPv6 service mesh — IPv6 service mesh is the architectural stance of giving every workload (Pod, VM, function, micro-VM) a unique IPv6 address on an encrypted private overlay,…
• IPv6 withdrawal as a shutdown leading indicator — A selective IPv6 address-space withdrawal from a country's in-country ASes — hours before a traffic drop is externally observed…
• IRR prefix list — An IRR prefix list is a list of BGP prefixes generated by expanding an AS-SET / as-macro object published in an Internet Routing Registry (IRR) — most commonly RIPE IRR, RADb,…
• Isolate vs microVM for agent sandbox — A design-axis trade-off for the substrate that runs an agent's execution sandbox — file ops, command exec, code execution, web browsing — distinguishing between:
• Isolated Recovery Environment (IRE) — An Isolated Recovery Environment (IRE) is a dedicated AWS account (or analogous cloud isolation boundary) used as the rebuild and validation surface for cyber-event recovery.…
• Isolation as fault-tolerance principle — Isolation as a reliability principle means systems are made from parts that are as physically and logically independent as possible,…
• Isomorphic observability API — An isomorphic observability API is an instrumentation interface that is runtime-neutral — written once, deployed unchanged on both the server (Node.js) and the browser…
• Iterative Coordinate Grid Probing — Iterative coordinate grid probing is a technique for extracting precise bounding-box coordinates from a multimodal LLM (VLM) on images where the model cannot reliably return…
• Iterative plan refinement — Iterative plan refinement is the agent-loop discipline where a Planner LLM produces a plan, a Coder LLM executes it, a Verifier LLM judges plan sufficiency,…
• Iterative prompt methodology — Iterative prompt methodology is the discipline of discovering a production prompt offline, through a sequence of discrete experiments,…
• Iterative prompt refinement — Iterative prompt refinement is the control-loop pattern where a generative model's output is scored against a rubric (typically by a judge LLM or VLM),…
• JavaScript heap size — JavaScript heap size is the memory the browser's JS engine (V8 in Chromium / Edge, SpiderMonkey in Firefox, JavaScriptCore in Safari) allocates for JS objects, closures,…
• jcmd thread dump — jcmd Thread.dumptofile is the Java 21 diagnostic command that captures a thread dump including the state of virtual threads.…
• JIT compilation — Just-in-time (JIT) compilation is the highest-performance point on the dynamic-language execution spectrum: at runtime, the engine compiles bytecode (or sometimes AST) directly…
• JIT peer provisioning — Just-in-time (JIT) peer provisioning is the architectural move of not installing a network peer in a kernel / data-plane primitive until the first packet from that peer arrives,…
• JNI transition overhead — JNI (Java Native Interface) is the JVM's foreign-function bridge to C/C++/assembly code. Each call crossing the JVM↔native boundary pays a non-trivial transition cost: the JIT…
• Join column type mismatch — A join column type mismatch occurs when the two columns on either side of a JOIN ... ON a.x = b.y predicate differ in type, size,…
• Join order optimization — Join-order optimization is the query-optimizer subproblem of choosing which order to execute a sequence of logical joins in.…
• JSON column as schema escape hatch — JSON column as schema escape hatch is the anti-pattern of using a MySQL JSON (or Postgres JSONB) column to absorb new fields as an alternative to running a schema migration.…
• JSON data type storage overhead — The storage-overhead caveat is the single largest reason to prefer classical relational columns over a MySQL JSON column when both shapes would work: the binary JSON encoding…
• JSON output stability — JSON output stability is the commitment that a CLI's structured output (JSON via --output json, YAML via --output yaml, or equivalent) uses field names that stay stable across tool…
• JSON-serializable DSL — A JSON-serializable DSL is a domain-specific language whose authoritative representation is a plain JSON (or structurally similar) data structure…
• JSON-string parameters for schema stability — JSON-string parameters for schema stability is the design choice to carry element-specific content as an opaque JSON string inside a stable, minimal outer schema,…
• JSONB streaming buffer — JSONB streaming buffer is the two-phase schema discipline where a streaming-ingest pipeline lands arriving records into a PostgreSQL JSONB column — schemaless ingest…
• JSONL output streaming — JSONL (JSON Lines) is a text format in which every line is a valid, self-contained JSON object. As a streaming protocol for long-running processes it resolves a specific pain:…
• jsonschema config validation — jsonschema config validation is the practice of declaring a single JSON Schema as the authoritative description of a configuration file's shape,…
• Jump table vs binary search dispatch — When a compiler lowers a switch statement with many integer cases, it has two main strategies:
• JSON Web Key (JWK) — A JSON Web Key (JWK) is a JSON representation of a cryptographic key — public, private, or symmetric — standardised by RFC 7517 as part of the JSON Object Signing and Encryption…
• Kafka consumer backlog — Kafka consumer backlog (a.k.a. consumer lag) is the offset delta between a partition's latest produced message and the consumer group's most recently committed offset…
• Kafka consumer group lag metric — Consumer group lag is the count of un-consumed messages in a Kafka partition: the difference between the log's latest offset (last produced record) and the consumer group's…
• Kafka offset+partition uniqueness constraint — Kafka offset+partition uniqueness constraint is a database-side idempotence primitive for at-least-once Kafka consumers: every row persisted from a Kafka message carries the source…
• Kafka partition — A Kafka partition is the unit of parallelism, ordering, and replication in Apache Kafka. Every topic is split into N partitions;…
• Keep hazmat away from complex code — A secure-design heuristic, stated by Fly.io's Thomas Ptacek as the explicit rationale for building tkdb:
• KeepAlive-message LSN advancement — KeepAlive-message LSN advancement is the technique of using Postgres logical-replication KeepAlive messages — periodic heartbeat frames sent server → client that carry the current…
• Kernel attack surface — Kernel attack surface is the set of kernel codepaths a compromised user-space process can reach, and therefore the set of kernel bugs it could weaponise.…
• Kernel Optimization as Search — Kernel optimization as search is the architectural primitive of reframing "pick the best GPU/accelerator kernel for this operator + hardware target" as a search problem over…
• Kernel panic from scale — A kernel panic from scale is the production failure mode where kernel code paths that work correctly at small / moderate state sizes trigger a panic once per-subsystem state grows…
• Kernel state capacity limit — Kernel state capacity limit is the observation that kernel subsystems that hold per-object state have a practical ceiling below which performance is fine and above which things…
• Key-overuse detection — Key-overuse detection is the operational practice of observing a cryptographic key's cumulative usage (bytes protected, operations performed,…
• Key partitioning — Key partitioning is a MySQL partitioning strategy that lets MySQL itself pick the partitioning function — an internal hash algorithm operating on the table's primary key (or…
• Keyed partitioner — A keyed partitioner is the Kafka / Redpanda producer-side partitioning strategy used when a record has a non-null key: the partition is chosen deterministically by partition =…
• Keyspace ID — A Keyspace ID is Vitess's shard-address identifier — a binary(8) value per row that tells VTGate which physical shard the row lives on.…
• Keyspace — A keyspace in Vitess is a logical database that holds a group of tables, optionally sharded, presenting a single SQL surface to the application.…
• kill / copy / boot migration tradeoff — The kill / copy / boot migration tradeoff is the classical dilemma that stateful-workload migration faces when relocating a workload anchored to a specific host's local storage.…
• Killswitch subsystem — A killswitch subsystem is a mechanism inside a rules engine, feature-flag system, or policy engine that allows a specific rule / flag / policy to be disabled rapidly,…
• Kinetic attack on cloud infrastructure — A kinetic attack on cloud infrastructure is the physical destruction or damage of hyperscaler data-center facilities by explosive, impact, or other kinetic-energy means — drones,…
• KLL quantile sketch — A KLL quantile sketch is a streaming probabilistic data structure that produces rank-error-bounded quantile / percentile estimates over a stream of numeric values,…
• Know Your Customer (KYC) — Know Your Customer (KYC) is the financial-services discipline of verifying customer identities, assessing risk profiles, and monitoring transactions for signs of money laundering,…
• Knowledge Distillation — Knowledge distillation is the technique of transferring knowledge from a large, capable, expensive-to-run teacher model to a small, fast,…
• Knowledge graph — A knowledge graph is a data structure that captures relationships between entities (people, documents, events, projects, activities) rather than just their individual contents.…
• Knowledge pyramid (model tiering) — A knowledge pyramid is a multi-agent architecture pattern that tiers model cost against task cognitive load: cheap models at the fan-out leaves where work is tool-call-heavy…
• Kotlin platform type — A Kotlin platform type is the type that the Kotlin compiler assigns to a value coming from Java when the Java side has no nullability annotation (@Nullable, @NonNull, @Nullsafe,…
• KPI-driven load ramp-up — KPI-driven load ramp-up is a load-test control strategy where the ramp schedule targets a business KPI (orders-per-minute, reservations-per-second,…
• KPU — Kinesis / Managed Flink Processing Unit — A KPU is AWS Managed Flink's unit of compute provisioning: a bundle of 1 vCPU + 4 GB memory + 50 GB running application storage.…
• KRaft metadata log (cluster metadata as a Kafka log) — The KRaft metadata log is Apache Kafka's post-ZooKeeper mechanism for storing cluster metadata. A special topic _clustermetadata (single partition) whose leader election runs via…
• KTLO engineering chores — KTLO — "keeping the lights on" — is the standard industry shorthand for maintenance work that doesn't move product forward but must be done to prevent the existing product…
• Kubernetes label length limit (63 characters) — Kubernetes metadata label values are limited to 63 characters. This comes from RFC 1123 DNS-label syntax that Kubernetes adopted for its label grammar.…
• Kubernetes Operator pattern — A Kubernetes Operator is a custom controller that extends the Kubernetes Control Loop (observe → diff → act) to manage a domain- specific workload…
• KV cache (transformer inference) — The KV cache is the per-layer, per-token Key and Value projection tensor store that a transformer decoder reuses across autoregressive generation steps.…
• L0/L1 file compaction for object-store streaming — L0/L1 file compaction is the two-tier object-storage file layout used by streaming systems that write primary record data directly to object storage: L0 (Level 0) files…
• Label-based log indexing — Label-based log indexing is the architectural choice to index only a small set of low-cardinality dimensions — labels — attached to log lines,…
• Lakebase Local File Cache (LFC) — The Local File Cache (LFC) is a Lakebase-specific caching layer on the compute-VM side of Lakebase that holds Postgres pages locally,…
• Lakehouse-native observability — Lakehouse-native observability is the architectural posture of storing observability data (metrics, logs, traces) as governed tables in a columnar-object-storage lakehouse (Delta…
• Lambda fan-out benchmark — A Lambda fan-out benchmark uses AWS Lambda's horizontal-concurrency scaling as the load generator for a downstream system.…
• Lameduck mode — Lameduck mode is a node state in which in-flight requests are allowed to complete but no new requests are accepted. It is the drain primitive under any graceful…
• Landing latency — Landing latency is the elapsed time between a source-data event (e.g. a row insert / update / delete in the source database) and the corresponding row appearing…
• Language extension vs replacement — Language extension vs replacement is the design question that shows up whenever an entrenched programming or query language has well-known design problems but its ecosystem (users…
• Larger-than-RAM vector index — A larger-than-RAM vector index is an ANN index whose total on-disk footprint exceeds the process's available memory, yet still serves queries at acceptable latency because only…
• Last-known-good routing table — Last-known-good routing table is the operational policy that a data-plane proxy continues serving traffic using the last routing configuration it successfully loaded even after its…
• Last-mile data processing — Last-mile data processing is the class of Python data work that happens after the heavy ETL has already landed data in the warehouse — feature transformations, batch inference,…
• Last Reconciled Offset — The Last Reconciled Offset is a per-partition watermark tracked by the Redpanda Cloud Topics architecture that marks the boundary between offsets whose data still lives only in L0…
• Last-write-wins — Last-write-wins (LWW) is a conflict-resolution rule for concurrent updates to the same state: if two replicas update the same field, the one with the later timestamp wins.
• Late-arriving data — Late-arriving data is event data that reaches the analytics / streaming system after the "wall clock" bucket that the event belongs to has already started being processed…
• Late Fusion (Multimodal) — Late fusion is the multimodal-ML strategy of training a separate model per modality and combining their predictions (by averaging, weighted voting, a meta-learner,…
• Latency-based DNS routing — Latency-based DNS routing is a DNS resolution policy in which a single hostname resolves to different IP addresses depending on the measured network latency from the resolver's…
• Latency-critical vs latency-tolerant workload — Latency-critical vs latency-tolerant workload is the workload- class distinction between streams whose business value depends on low end-to-end latency (tight p99 / p99.9 targets)…
• Latency rises before throughput ceiling — In a load-tested system under increasing concurrency, client-observed p99 latency rises before aggregate throughput outright plateaus.…
• Latent-diffusion video generation — Latent-diffusion video generation is an approach to generative video where the diffusion process — iteratively denoising from Gaussian noise toward a coherent output…
• Latent misconfiguration — Latent misconfiguration is a configuration bug that is structurally wrong from the moment it lands in production but produces no observable effect until some later,…
• Lattice-based cryptography — Lattice-based cryptography builds cryptographic primitives — encryption, signatures, key exchange, homomorphic encryption,…
• Layer 7 load balancing — Layer 7 load balancing makes routing decisions at the application protocol layer (HTTP headers, gRPC method, request path, query parameters) — and, critically,…
• Layered abstraction as human crutch — A claim articulated in the 2026-02-24 vinext launch:
• Layered grounded context for data agent — A design pattern for AI data agents (LLM agents that turn natural-language questions into SQL against a real data warehouse) in which multiple distinct layers of grounded context…
• Layered protection infrastructure — Layered protection infrastructure is the design posture of distributing defense mechanisms across multiple ordered tiers of a request-serving stack — typically edge, application,…
• Layering violation — A layering violation occurs when a component implemented at one layer of the system reaches into the responsibilities of another layer…
• Lazy container image loading — Lazy container image loading is the technique of starting a container before its filesystem has been fully fetched, and streaming the filesystem layers on demand as the container's…
• Lazy history download — Lazy history download is the VCS design stance in which clone downloads almost no history (commits, trees, files) and instead pulls data on demand as the user runs queries…
• Lazy Hydration — Lazy hydration is an initialization pattern: on first exposure of a data view to an application, populate the view's metadata eagerly (so operations work immediately) but defer…
• Lazy leaky bucket update — A lazy leaky bucket data structure avoids a dedicated background thread to drain buckets or evict idle buckets. Instead, both operations are amortised into the arrival path:
• Lazy-loaded BIOS data structure — A lazy-loaded BIOS data structure is a UEFI/BIOS internal structure (typically an HII / IFR data structure describing configuration forms) whose contents are not instantiated…
• Lazy-loading container filesystem — A lazy-loading container filesystem is a container-image delivery model that does not download the whole image before the container can start.…
• Leader establishment — Leader establishment is the act of satisfying the conditions required for a node to act as leader — i.e., to successfully complete requests on behalf of the group.…
• Leader-follower replication (Kafka partition) — In Apache Kafka, every partition has exactly one leader at a time; the other replicas are followers. Writes always go to the leader;…
• Leader lease — A leader lease is a time-bounded grant to the current leader during which no leadership change will occur. While the lease is valid,…
• Leader pinning — Leader pinning is the mechanism by which an operator specifies preferred locations (regions, racks, or AZs) for the partition leaders of a topic on a multi-region or multi-AZ…
• Leader revocation — Leader revocation is the act of invalidating the conditions that allow a leader in a single-leader distributed system to complete requests.…
• Leaked-credential auto-invalidation — Leaked-credential auto-invalidation is the safety primitive where a platform that issues credentials for its own service automatically revokes any credential that is detected…
• Learned lifetime distribution — Learned lifetime distribution is the prediction unit of emitting a probability distribution over remaining lifetime rather than a point estimate.…
• Least-privileged access — Each actor — user, service, operator — sees only the data and gets only the capabilities strictly required for their current interaction.…
• Leftmost prefix rule — The leftmost-prefix rule states that a composite index on columns (A, B, C, …) can answer only those queries whose WHERE clause filters on a contiguous leading prefix…
• Legacy-format emission for incremental sunset — Legacy-format emission for incremental sunset is the technique of having a new canonical service re-emit the old legacy data formats — on the original transport…
• Legacy native module main-queue deadlock — Legacy native module main-queue deadlock is a failure mode in React Native on iOS where a legacy (bridge-era, non- TurboModule) native module that has declared…
• Legacy shared account — A legacy shared account is a login on a third-party platform that sits outside the org's primary IdP-gated regime — typically because the platform doesn't support SSO federation…
• Legacy system risk — Legacy system risk is the elevated-failure-probability surface that accumulates when production systems cannot adopt modern delivery practices — automated testing,…
• Less prompting is more for analytical agents — A design lesson for LLM agents whose task is analytical reasoning over a known data substrate (writing SQL, navigating a metadata catalog,…
• LFRU cache eviction (Least Frequently Recently Used) — Least Frequently Recently Used (LFRU) is a two-queue cache replacement policy that gives frequently-accessed items a second chance before eviction.…
• Lift metric — In an interleaving test of ranking A vs ranking B, the lift metric aggregates per-search (or per-user) preference into a single scalar.
• Lightweight formal verification — Lightweight formal verification is a family of techniques that sit between ad-hoc testing and heavyweight proof-based formal methods (TLA+, Coq, Isabelle).…
• Light-weight rollup event — A light-weight rollup event is a notification emitted from the write path of an event-log-based aggregation system that carries no payload…
• LIKE predicate cardinality estimation failure — LIKE-predicate cardinality estimation failure is the canonical blind spot of histogram-based query-planner cardinality estimators: string-pattern predicates (LIKE '%foo%',…
• Linear probe evaluation — Linear probe evaluation is a standard evaluation methodology for self-supervised foundation models where:
• Linear-search timeout amplification — Linear-search timeout amplification is the failure mode in which a system, looking up a value through a sequence of candidate options with a per-option timeout for unreachable…
• Linear shard-count throughput scaling — Linear shard-count throughput scaling is the empirical property that doubling the shard count approximately doubles the QPS ceiling of a horizontally-sharded cluster,…
• Linear vs super-linear cost scaling — Linear cost scaling: when workload demand grows by factor N, cloud cost also grows by factor N.
• Link discovery vs link value assessment — Googlebot's crawl-to-index pipeline splits link-handling into two distinct stages that happen at different points and on different inputs:
• Link response header (RFC 8288) — Link: is an HTTP response header standardised as RFC 8288. Each value carries a URL plus a rel="..." relation type, expressing a typed link from the current resource to another…
• Link-state routing protocol — A link-state routing protocol is a class of network routing protocols (OSPF, IS-IS) where each router is the source of truth for its own links and is responsible for quickly…
• Linux cGroup — A Linux cGroup (control group) is a kernel primitive that groups a set of processes and lets the kernel enforce resource limits (CPU, memory, I/O,…
• Linux namespaces — Linux namespaces are the kernel primitive that isolates what a process can see of the system — a process in a namespace sees its own view of a resource type (processes,…
• Linux page cache — Linux page cache is the kernel's in-memory cache of file-system pages; buffer cache is its sibling for block-device I/O (often merged into page cache on modern kernels).…
• Liquid-cooled AI rack — A liquid-cooled AI rack is a rack-scale compute unit where heat from the GPUs/accelerators is removed via a liquid coolant loop (water, dielectric fluid,…
• List partitioning — List partitioning is a MySQL partitioning strategy that assigns rows to partitions based on membership in a fixed, pre-declared set of discrete values for the partition-key column.…
• LISTEN/NOTIFY payload limit — Postgres's built-in LISTEN / NOTIFY pub/sub mechanism enforces an 8 KB payload limit on each NOTIFY message. Any application wanting to fan out larger messages via this channel…
• Little's Law — Little's Law is the foundational result in queueing theory due to John Little (1961, MIT) that states: in any stable queueing system,…
• Live-off-the-land — Live-off-the-land is the architectural posture of accomplishing a task on a remote host using only the tools already present there — no agent deployed, no runtime shipped,…
• LLM agent as query optimizer — LLM-agent-as-query-optimizer is the architectural pattern of applying a frontier-model LLM agent, given one or more database execution tools and a bounded rollout budget,…
• LLM as Judge — LLM-as-judge is the evaluation pattern in which one LLM scores another model's (or agent's) output against a rubric — accuracy, helpfulness, policy adherence,…
• LLM-based cluster evaluation — LLM-based cluster evaluation uses an LLM as a judge to score each leaf group in a discrete-code (e.g. Semantic ID) hierarchy on multiple business-meaningful dimensions…
• LLM-based ranker — An LLM-based ranker is an architectural role in which a large language model scores or orders items from a finite candidate set rather than generating unconstrained text.…
• LLM batch API — An LLM batch API is a provider-side inference endpoint that accepts a bulk file of prompts (typically JSONL) and returns bulk results asynchronously,…
• LLM cascade — An LLM cascade is a cost-vs-quality routing pattern in which the cheapest adequate LLM runs first, and a progressively more expensive,…
• LLM code-generation error rate — The LLM code-generation error rate is the fraction of code generations from a language model that fail to produce a working artifact (a compiling program, a rendering website,…
• LLM conversion hallucination control — LLM conversion hallucination control is the structural problem class that arises when an LLM is asked to perform a deterministic code-transformation task (e.g.…
• LLM decoding step — The decoding step is the final phase of LLM text generation where the model transforms its internal representations into human-readable tokens. At each step the model:
• LLM evaluation dimensions — LLM evaluation dimensions are the top-level axes of a rubric used to score an LLM-powered system's output quality — each dimension decomposes into granular per-criterion True/False…
• LLM-generated database changes — LLM-generated database changes is the class of workflows where a large language model is asked to propose schema-level mutations (indexes, constraints, column additions,…
• LLM-generated prompt regression test — An LLM-generated prompt regression test is a CI-level test that runs the production LLM pipeline against a checked-in library of LLM-generated,…
• LLM hallucination — LLM hallucination is the failure mode where a language model "confidently makes claims that are incorrect" — it generates output that is linguistically plausible…
• LLM hyperlink hallucination — LLM hyperlink hallucination is the failure mode where a language model fabricates URLs in its output — emitting plausible-looking,…
• LLM icon hallucination — LLM icon hallucination is a specific code-generation failure mode where a language model references icons from an icon library "that no longer exist or never existed."…
• LLM model drift — LLM model drift is the failure mode where a deployed language model's behaviour changes over time — producing different (and often worse) outputs for the same prompt after weeks…
• LLM model feature lag — LLM model feature lag is the time delay between when a model provider (Anthropic, OpenAI, Google, Meta, etc.) makes a new model version, optimisation,…
• LLM over-provisioning cycle — The LLM over-provisioning cycle is the structural failure mode of static / dedicated LLM serving capacity (Provisioned Throughput on Bedrock or equivalent) when workload demand…
• LLM provider commitment lock-in — LLM provider commitment lock-in is the model-upgrade friction introduced by multi-month dedicated-capacity contracts on managed LLM serving substrates (Amazon Bedrock Provisioned…
• LLM segmentation over traditional NER — LLM segmentation over traditional NER is the framing that segmentation tasks (assign labels to consecutive token-runs of a query) are structurally better served by a prompted large…
• LLM self-verification — LLM self-verification is the technique of obtaining a confidence score for an LLM's output by asking the same (or a second) LLM a follow-up entailment question…
• LLM training-data exhaustion — LLM training-data exhaustion is the claim — formalised by Epoch AI (2024) — that frontier language models are approaching the upper bound of the publicly-available human-generated…
• LLM workflow router — An LLM workflow router is an architectural primitive where a language model classifies an inbound user request into one of N pre-defined workflows,…
• llms.txt — llms.txt is a plain-text file at the root of a website (e.g. https://example.com/llms.txt) that gives an LLM agent a structured reading list: what the site is, what's on it,…
• LLVM BOLT post-link binary optimizer — LLVM BOLT (Binary Optimization and Layout Tool) is a post-link binary optimiser — it consumes an already-compiled, already-linked executable plus a profile data file and rewrites…
• Load average — Load average on Linux is a set of three numbers exposed by the uptime command and several others (top, w, /proc/loadavg): exponentially damped moving averages of the number…
• Load shedding at ingestion — Load shedding at ingestion is the architectural principle of refusing (or slowing) work at the boundary where it enters the system,…
• Loadable MySQL function — A loadable MySQL function (often called a user-defined function / UDF in MySQL terminology, or "loadable function" in 8.0+ vocabulary) is a native C or C++ function compiled into…
• Local emulation — Local emulation = running a faithful-enough substitute for a cloud service on the developer's local machine so that application code targeting the real service's API can be…
• Local-first architecture — Local-first architecture is a system-design posture in which all state, all processing, and all control live on user-owned hardware — not on a vendor's cloud…
• Local-global aggregation decomposition — Local-global aggregation decomposition is the algebraic split of an aggregate function into two parts:
• Local MCP server risk — Local MCP server risk is the security posture a developer takes on when they wire a cloud-hosted LLM (Claude, Cursor, GitHub Copilot, Goose,…
• Local rate-limit decision — A local rate-limit decision is an architectural choice: the request-time "accept or reject" call happens inside the requesting host's process (via an embedded library reading…
• Local-remote parity — Local-remote parity is the design goal of making local development mirror the production cloud API at the API shape level, not just at the runtime-behavior level.…
• Locale / host-default SSR/CSR divergence — A class of hydration mismatches where the same Intl. or toLocaleString call produces different output on the server than on the client because the API defaults to the host's…
• Locality-aware scheduling — Locality-aware scheduling is the scheduler policy of placing a unit of work on the node that already holds its input data, minimising network transfer.…
• Lock contention in query planning — Lock contention in query planning is the failure class where a database's per-cluster metadata lock — typically protecting the table's list of physical storage units (parts,…
• Lock-state self-synchronizing update — A self-synchronizing atomic update is the optimisation of clearing specific bits of a packed state word not by load-modify-CAS but by atomically adding a specific delta such that…
• Lock timeout hedging — Lock timeout hedging is the policy choice, in a blocking-lock protocol, of "wait for the lock up to a bounded window; if it doesn't come,…
• Locking read (MySQL) — A locking read in MySQL is a SELECT statement that acquires row-level locks on the rows it returns, promoting it from a plain non-locking read to a transactional primitive…
• The log is the truth, the database is a cache — The truth is the log. The database is a cache of a subset of the log. A framing — originating in Kleppmann's CIDR 2015 paper "Turning the database inside out"…
• Log recovery time (broker disk rebuild) — When a Kafka broker restarts after an ungraceful shutdown, it must rebuild the local log-index files associated with each partition replica it hosts.…
• Logging mode vs enforcement mode — In Meta's Policy Zones (and any IFC-style runtime enforcement system), a zone operates in one of two modes:
• Logical component grouping for context budget — Logical component grouping for context budget is the discipline of partitioning a large transformation corpus into small,…
• Logical data model — A logical data model is "a set of definitions of tables and columns in which the consolidated record pulled, matched, and merged from the different sources is stored" ().
• Logical replication as pub/sub — Logical replication as pub/sub is the architectural observation that Postgres logical replication — normally deployed for cross-database replication or CDC…
• Logical replication — Logical replication is a replication mode in which the primary database emits a stream of row-level change events (insert / update / delete with primary-key and column values),…
• Logical vs physical backup — Database backups fall into two architectural families:
• Logical vs physical sharding — Logical sharding and physical sharding are two different states in the horizontal-sharding rollout. Decoupling them — doing logical before physical,…
• Logit verifier for model porting — A logit verifier is a mechanically-checkable correctness criterion used when porting a new model family from a reference implementation (typically Hugging Face transformers) into…
• Logits (LLM / transformer) — Logits are the pre-softmax prediction scores a transformer LLM emits over the vocabulary at each generation step. For a vocabulary of size V,…
• Logless Reconfiguration — Logless reconfiguration is MongoDB's replica-set membership-change protocol that decouples configuration changes from the data replication log.…
• Long-context reasoning — Long-context reasoning is the cluster of LLM capability problems that arise as conversations / inputs / RL trajectories grow long.…
• Long Fork anomaly — The Long Fork anomaly is a violation of Snapshot Isolation's atomic-visibility property in which *two readers on different nodes (or at different points in time) observe two…
• Long-lived key risk — Long-lived key risk is the principle that the impact of a compromised key scales with the duration the key grants access and the breadth of systems that trust it — so migration,…
• Long-tail query — A long-tail query is a user search query that appears rarely or never in historical traffic — highly specific, uncommon, or creatively phrased.…
• Long user sequence modeling — Long user sequence modeling is the use of a Transformer encoder over a long history of user actions (viewed/liked/engaged items, queries, dwells,…
• Lookup sharding — Lookup sharding routes each row to a shard via a developer-maintained mapping table that the proxy consults on each query…
• LoRA (Low-Rank Adaptation) — LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning (PEFT) technique that freezes the weights of a pre-trained base model and trains only a small number of new…
• Loss masking assistant tokens — Loss masking in SFT is the practice of explicitly marking which tokens in a training example contribute to the loss and which don't.…
• Lossless weight compression — Lossless weight compression is the problem class of reducing an LLM's on-device weight footprint so every weight reconstructs bit-exactly to its original value…
• Lost in the middle effect — The "lost in the middle" effect is a named failure mode of long-context LLM prompting where "details in the middle of long inputs are often overlooked or distorted"…
• Low-bit inference — Low-bit inference is the umbrella practice of serving attention-based neural networks with sub-FP16 numerical precision for activations and/or weights — 8-bit, 4-bit,…
• Low-cardinality clustering optimization — The low-cardinality clustering optimization is the layout specialisation Liquid Clustering applies when one of the clustering keys has low cardinality: each data file is laid out…
• Low-end device inclusion — Low-end device inclusion is the product-level constraint that engineering choices — codec, ML model, rendering pipeline — must serve users on older, slower, cheaper devices,…
• LRU cache eviction — Least Recently Used (LRU) is the industry-standard cache replacement policy. When the cache is full and a new entry needs a slot,…
• LSM compaction (size-tiered, leveled, hybrid) — Log-Structured Merge (LSM) is the standard storage organisation for write-heavy systems that need to serve reasonably fast point / range / scan queries over the same data.…
• lsn.flush.mode (Debezium) — lsn.flush.mode is a Debezium Postgres connector configuration property introduced in Debezium 3.4.0.Final (2025-12-16) that controls who is allowed to advance the Postgres logical…
• LTS kernel backport latency gap — The LTS kernel backport latency gap is the structural window between when a Linux kernel fix lands in mainline and when it reaches the Long-Term Support (LTS) series…
• LTX file format (LiteFS Transactions) — LTX — LiteFS Transactions — is the on-wire / on-disk file format Fly.io's LiteFS uses to represent sorted, per-transaction changesets of SQLite pages.…
• LTX index trailer — The LTX index trailer is a small, fixed-format section at the tail of each LTX file containing an index that maps page number → byte offset (and size) inside the file.…
• M:N scheduler — An M:N scheduler multiplexes M userspace tasks (green threads, fibers, goroutines, actors) onto N kernel threads where typically M >> N.…
• M2M OAuth vs PAT — Machine-to-machine (M2M) OAuth versus Personal Access Tokens (PATs) is the auth-substrate choice for service-to-service authentication: which credential class does a non-human…
• Macaroon token — Macaroon tokens are bearer tokens (like JWTs) that use a chained-HMAC construction allowing any holder to take an existing token and scope it down…
• Machine metadata service — A machine metadata service is a guest-local endpoint that a workload running inside a VM / container / Machine uses to obtain information about itself from the platform that runs…
• Machine payment — A machine payment is a payment made by an agent directly to a service, programmatically, during task execution — without a human buyer's checkout moment.…
• Machine-readable command catalog — A machine-readable command catalog is a parseable inventory of a CLI's entire surface — every command, every flag, every type,…
• Machine-readable documentation — Machine-readable documentation = repository docs structured for AI agents first, humans second: short, factual, consistently-named files with predictable layout — AGENT.md,…
• Machine translation with LLMs — Machine translation with LLMs (MT-via-LLM) is the use of general-purpose large language models as the translation engine for production localization systems,…
• Maintenance domain — A maintenance domain is the fraction of fleet capacity taken offline together in a single maintenance action. It is the unit of blast radius for a routine fleet operation (firmware…
• Maintenance window — A maintenance window is a customer-configurable contract with a managed-service vendor specifying when the vendor is allowed to perform disruptive operations — routine patches,…
• Major-version upgrade discipline — Major-version upgrade discipline is the set of practices a team applies when upgrading a stateful datastore across a major-version boundary,…
• Majority-consensus replication — Majority-consensus replication is the replication scheme in which a write becomes durable once a strict majority of replicas acknowledge it.…
• Managed data plane — Managed data plane is the architectural primitive where the vendor operates the data-plane processes, not just the control plane.…
• Managed Kubernetes service — A managed Kubernetes service is a hosted offering where the cloud provider operates the Kubernetes control plane — and, in stronger variants,…
• Manifest replication via Raft — In an LSM-tree database, the manifest tracks which SSTables exist and their key ranges. Traditionally it's a local file rewritten on each flush.…
• Map-fold LLM pipeline — A map-fold LLM pipeline is a functional-composition pattern for processing a large document corpus through LLMs: a map phase applies a per-document LLM invocation to extract…
• Markdown as agent-friendly format — Markdown as agent-friendly format is the structural property of Markdown — line-per-record, natural column alignment, grep-friendly prefixes,…
• Markdown content negotiation — Markdown content negotiation is the convention that when a client sends Accept: text/markdown on an HTTP request, the server responds with a clean markdown representation…
• Markdown sitemap — A markdown sitemap is a site's URL-index served at a well-known path (e.g. /blog/sitemap.md, /docs/sitemap.md) as a hierarchical markdown table of contents instead of a flat XML…
• Market Group (country-level isolation) — A Market Group is a named deployment instance of a serving API that owns a subset of the geographic markets (countries) the platform serves.…
• Market-mediated long-term effects — In a multi-sided marketplace (riders + drivers, buyers + sellers, hosts + guests), an intervention on one side propagates through shared market state — prices, wait times,…
• Masked Shot Modeling (MSM) — Masked Shot Modeling (MSM) is a self-supervised pre-training objective in which a sequence model is trained to predict the original embedding of masked shots in a video,…
• Master data management — Master Data Management (MDM) is "a technology-enabled discipline in which business and Information Technology work together to ensure the uniformity, accuracy, stewardship,…
• Master-node resource contention — Master-node resource contention is the operational failure mode in which jobs orchestrated against a managed Hadoop cluster (EMR / on-prem ResourceManager) end up executing…
• Master Patient Index (MPI) — A Master Patient Index (MPI) — sometimes called MDM (Master Data Management) for patient records — is the deduplication primitive that ensures one patient = one golden record…
• Materialized view — A materialized view (MV) is a database object that persists the result of a query as a physical table. Unlike a regular view…
• Matrix multiply-accumulate (MMA) — Matrix multiply-accumulate (MMA) is the fused primitive C ← A × B + C at fixed hardware-defined tile sizes, exposed by NVIDIA Tensor Cores and AMD Matrix Cores via dedicated…
• max_connections ceiling — maxconnections is MySQL's server-side configuration variable capping the number of concurrent client connections the server will accept.…
• Max-min fairness storage fair-share — Max-min fairness storage fair-share is the allocation policy in which a shared multi-tenant storage budget is divided among tenants such that the smallest allocation is maximised…
• MCP client-config fragmentation — MCP client-config fragmentation is the observation that every MCP-aware client (Claude Desktop, Cursor, Zed, VS Code, Windsurf, Neovim,…
• MCP long-lived SSE — A routing / runtime property of modern Model Context Protocol (MCP) deployments: MCP flows between an LLM client and a remote MCP server now include repeated,…
• MCP registry — An MCP registry is an organisation-internal authoritative catalog of approved MCP servers — the source of truth for which servers are allowed to serve production traffic at a given…
• MCP resource — An MCP resource is a read-only, addressable, client-cacheable, optionally-subscribable piece of content exposed by a Model Context Protocol server.…
• MCP Server Card — MCP Server Card is a draft well-known JSON document that describes a Model Context Protocol server before an agent connects to it.…
• MCP tool vs resource — MCP tool vs resource is the decision rule inside the Model Context Protocol for whether to expose a capability as a tool (invocable action with side effects) or a resource…
• Mean Reciprocal Rank (MRR) — Mean Reciprocal Rank (MRR) is a statistical metric for evaluating any process that produces a ranked list of candidate responses to a query,…
• Medallion Architecture — The Medallion Architecture is a three-tier pattern for organising data inside a data lakehouse, structured by progressive data quality. Each tier is a named "colour" of refinement:
• Memory-aware scheduling — Memory-aware scheduling is the scheduler policy of sizing and placing work against memory as a first-class resource, not CPU alone.…
• Memory-bandwidth-bound inference — Memory-bandwidth-bound inference is the regime in which per-token latency is gated by bytes moved from HBM to the compute units, not by FLOPs executed once the data arrives.…
• Memory-bound vs compute-bound (GPU inference) — A workload is memory-bound when its performance is limited by how fast data can be moved between GPU memory (HBM) and the compute units…
• Memory compaction — Memory compaction (also called context compaction) is the lifecycle moment in an agent loop at which the context window is shortened…
• MemoryOffsetBackingStore (Debezium) — MemoryOffsetBackingStore is a Kafka Connect offset-store implementation used by Debezium (and embeddable via Debezium Engine) that holds connector offsets in JVM heap memory only…
• Memory overcommit risk — Memory overcommit risk is the failure mode where a standalone database's configured maxconnections exceeds the worst-case aggregate memory demand across all simultaneously active…
• Memory profiling granularity — Memory profiling granularity is the set of keys a memory accounting system aggregates allocations by. MySQL's performanceschema exposes five fixed granularities — account, host,…
• Memory safety — Memory safety is the property that a program cannot access memory it isn't authorized to — no use-after-free, no buffer overrun, no double-free, no uninitialized-read.…
• Memory supersession — Memory supersession is the discipline of updating an agent-memory fact or instruction by keeping the old memory, linking it forward to the new one,…
• Mental-model preference coherence — Mental-model preference coherence is the architectural discipline of designing preference schemas so that the storage model matches the user's internal model of what the system…
• Merchant delinquency risk — Merchant delinquency risk is the predictive primitive that estimates whether a business on a payments platform is likely to accrue a negative balance and have that balance remain…
• Merge-Base Three-Tree Sync — Merge-base three-tree sync is a file-sync data model that persists three observed tree states — Remote, Local, and a merge base called Synced…
• Merge-on-read (MOR) — Merge-on-Read (MOR) is the row-level update strategy in open table formats where mutations are persisted as separate delta files (append-only insert / update / delete records)…
• Merge queue — A merge queue is a queue of accepted pull requests that are validated against the would-be-future state of the target branch before being merged.…
• Mergeable sketch — A mergeable sketch is a probabilistic data structure with a merge operator that is associative and commutative, and whose merge produces a new sketch of the same type with error…
• Merkle tree — A Merkle tree is a binary tree of hashes in which every internal node is the hash of its two children's hashes. Named for Ralph Merkle's 1979 construction.…
• Mesh latency health-check — Mesh latency health-check is the pattern of having every edge node in a POP mesh continuously ping every other edge node as part of routine health-checking,…
• Messaging platform 24-hour response window — A platform-imposed business-messaging rule in which a bot account may send free-form outbound messages to a user only within a fixed window after the user's last inbound message.…
• Meta-monitoring — Meta-monitoring is a dedicated layer of monitoring whose sole job is to watch the monitoring stack itself — answering "is the system that would alert us if something is broken…
• Metadata boost — Metadata boost is a query-time ranking adjustment that nudges results up or down based on document metadata values (recency, priority, region, language, tenant,…
• Metadata/data-split storage — A storage-system architectural shape: split persistent state into two layers that are operated and scaled independently:
• Metadata-only embedding — Metadata-only embedding is the RAG vectorstore construction strategy where each document contributes multiple separately-embedded segments derived from its metadata…
• Metadata-only operation — A metadata-only operation is a SQL operation whose answer can be computed entirely from the table's metadata layer — typically the per-file min/max/null-count statistics…
• Metaflow extension mechanism — The Metaflow extension mechanism is the published (but not-yet- stable) API that lets organisations plug company-specific integrations into Metaflow — new compute backends,…
• Metamodel — A metamodel is a model of models — a formal specification whose instances are themselves modelling languages or domain models.…
• Metaprogramming on broken code — Metaprogramming on broken code is the practice of using AST-level tooling that can parse, analyse, and rewrite source that does not currently compile.…
• Metric aggregation as cardinality shield — Metric aggregation is the transformation of raw high-cardinality metrics into lower-cardinality aggregated series by dropping or collapsing expensive labels during ingestion.…
• Metric cardinality — Metric cardinality is the number of unique combinations of label values a metric has — e.g., cpuusage{pod="...", tenant="..."} has one series per distinct (pod, tenant) pair.…
• Metric definition as code — Metric definition as code is the practice of treating business metric definitions — the SQL logic, allowed dimensions, time granularities,…
• Metric definition drift — Metric definition drift is the anti-pattern where different teams independently maintain their own SQL/logic for the same business metric,…
• Metric-granularity mismatch — Metric-granularity mismatch is the observability failure mode where a dashboard or integration surfaces a metric at the wrong aggregation level for the question the operator…
• Metric sampling interval — Metric sampling interval is the period between successive measurements of a metric. A metric captured once per second has a sampling interval of 1 s;…
• Metric staleness from polling layers — When a metric travels through multiple independent polling stages (agent collects from source; consumer polls agent), the worst-case age of the metric seen by the consumer…
• Metric temporality (delta vs cumulative) — Temporality is how a metric data point expresses change over time:
• Metric-type metadata — An explicit, authoritative record of each metric's type (counter / gauge / histogram / summary) and unit, stored as metadata rather than inferred from naming conventions.…
• Metric view materialization — Metric view materialization is the substrate property where the data platform automatically maintains pre-aggregated results behind a headless-BI semantic layer metric definition,…
• Metrics tiering by criticality — Metrics tiering by criticality is the observability-layer discipline of classifying every emitted metric into a priority tier (typically three tiers) so that an overloaded metrics…
• MFU (Model FLOPS Utilization) — MFU (Model FLOPS Utilization) is a GPU-training efficiency metric: the ratio of observed model FLOPS achieved to the theoretical peak FLOPS of the hardware.…
• Micro-batching — Micro-batching is a stream-processing execution model that groups incoming records into small, time-bounded batches (typically 100 ms – a few seconds) and processes each batch…
• Micro-frontends — Micro-frontends is the application of microservice-style ownership to the frontend: a page or application is composed at runtime from independently owned, independently developed,…
• Micro-VM as Pod — Micro-VM as Pod is the architectural stance of making the Kubernetes Pod a thin compatibility layer over a micro-VM (typically Firecracker) rather than over a Linux container…
• Micro-VM Isolation — Micro-VM isolation is the practice of running each tenant's code inside a minimal hardware-virtualised VM (fast boot, small memory overhead),…
• Microbenchmark-vs-end-to-end gap — The microbenchmark-vs-end-to-end gap is the gulf between an optimisation's measured improvement in a narrow isolated benchmark (a single function's loop) and the same…
• Microservices migration — A microservices migration is the organisational and architectural transition from a monolithic codebase + shared database deployment model to many independently-deployed services…
• Microtask hop cost — Microtask hop cost is the per-invocation CPU + memory overhead of resolving a Promise through JavaScript's microtask queue — even when the resolution value is already available.…
• Migration job lifecycle — A migration job lifecycle is the per-job state machine governing where a job is in a multi-phase migration between two parallel implementations of the same logical pipeline.…
• Migration tracking table — A migration tracking table is a database table, inside the managed schema itself, where a versioned schema migration tool records which migration scripts have already been applied.…
• MIME type as first-class metadata — MIME type as first-class metadata is the API-design discipline of always declaring a MIME type on every piece of addressable content returned from a service or protocol endpoint…
• Min-cost flow — Min-cost flow is a class of graph algorithms that find the lowest-cost way to push a specified amount of flow from a source node to a sink node through a directed graph where each…
• Minimum-cwnd death spiral — A minimum-cwnd death spiral is a self-perpetuating failure mode in a loss-based congestion controller where, after a severe congestion event has collapsed cwnd to its minimum…
• Minority-quorum writeability — Minority-quorum writeability is the structural property of a replication system in which a strict minority of the nodes can still satisfy the write-acknowledgement rule…
• MirrorMaker2 async replication — MirrorMaker 2 (MM2) is Apache Kafka's canonical cross-cluster asynchronous replication tool — a Kafka Connect- based application that consumes from a source cluster and produces…
• Missing Modality Problem — The missing-modality problem is the production failure mode where a multimodal model — trained on examples where all modalities are present…
• Mixed-version cluster — A mixed-version cluster is the steady-state a cluster is in during a rolling upgrade: some nodes run the old version, some run the new version,…
• Mixed-workload contention — Mixed-workload contention is the pathology that arises when a single database tries to serve both OLTP and OLAP shapes from the same physical resources (CPU, memory,…
• Mixture of Experts (MoE / MMoE) — Mixture of Experts (MoE) is the neural network architecture pattern where a set of specialist subnetworks ("experts") process inputs,…
• ML bot fingerprinting — ML bot fingerprinting is the use of machine-learning classifiers over content-independent request features — TLS handshake fingerprints (JA3 / JA4), HTTP/2 frame ordering,…
• ML-first architecture — ML-first architecture is a chip-design posture that reverses the traditional scalar-first precedence: instead of starting from a general-purpose CPU / microcontroller and bolting…
• ML governance at scale — ML governance at scale is the architectural discipline of embedding compliance, access control, and data classification enforcement directly into the ML execution layer…
• ML platform architecture — An ML platform architecture is the set of infrastructure design decisions that govern how an organization standardizes the full ML lifecycle…
• ML Stack Tax — The ML Stack Tax is the operational burden of manually re-tuning serving infrastructure every time an organization deploys a new model or its traffic patterns shift.…
• Mock object maintenance cost — Mock object maintenance cost is the load-bearing but usually- unaccounted engineering cost associated with keeping test doubles (mocks, stubs,…
• Modality Masking During Training — Modality masking during training is the discipline of randomly removing one or more modality inputs from each training example so the model learns to predict without them.…
• Model-agnostic ML platform — A model-agnostic ML platform is a shared internal service that abstracts over a specific class of models (LLMs, image generators, embedding models, speech models,…
• Model-agnostic orchestration — The design principle of treating LLMs as interchangeable components within an orchestration pipeline rather than building tightly around a single model.…
• Model-aware vertical scaling — Model-aware vertical scaling adjusts how many concurrent requests each pod accepts (target_concurrency) based on the model's actual resource profile under load…
• Model bias toward finding something — Model bias toward finding something is the empirically observed tendency of LLMs, when asked to perform an exploratory task ("find bugs in this code",…
• Model drift monitoring — Model drift monitoring is the practice of continuously computing statistical metrics about model performance on fresh data and alerting when those metrics cross thresholds…
• Model Ensembling for Detection — Model ensembling for detection is the practice of running multiple detection models on the same input and combining their outputs into a single consensus detection set,…
• Model FLOPs Utilization (MFU) — Model FLOPs Utilization (MFU) is the ratio of FLOPs actually performed by a model's training or inference run to the peak theoretical FLOPs of the hardware over the same wall-clock…
• Model organic refusal inconsistency — Model organic refusal inconsistency is the failure mode where a frontier LLM's emergent (unprogrammed) refusals to sensitive requests are real but not reproducible…
• Model-per-Endpoint Isolation Trade-off — A production-ML serving decision: give each model its own endpoint and its own instance(s) (one-model-per-endpoint) versus multiplexing many models onto shared instances…
• Model serving vs model inference — Netflix explicitly distinguishes model serving from model inference in its ML-platform architecture. The distinction is small in sentence form but load-bearing for the platform's…
• Model signature as source of truth — Model signature as source of truth is the design discipline that an ML model's input / output declaration — exported as a first-class artefact alongside the model weights…
• Send the model to the data (not data to the model) — Send the model to the data is the enterprise-AI deployment thesis that for the bulk of an enterprise's data — the private data that constitutes its competitive moat…
• Model-to-feature binding — Model-to-feature binding is the design discipline of matching the specific latent strengths of a model to the specific requirements of a feature,…
• Model unit utilization ratio — The model unit utilization ratio is the autoscaling signal Databricks uses for LLM serving:
• Model units (LLM request cost abstraction) — A model unit (MU) is a Databricks-coined unit-of-account that quantifies the multi-dimensional cost of an LLM serving request as a single scalar load metric.…
• Module Federation host / remote topology — Host / remote is the runtime bundle topology introduced by Webpack Module Federation:
• MongoDB change streams — MongoDB change streams are MongoDB's native CDC mechanism exposing a cursor over the replica set's operations log (oplog)…
• Monitoring paradox — Monitoring paradox: the observability layer deployed to catch infrastructure problems becomes an infrastructure problem. Werner Vogels's phrasing:
• Monolith vs microservices pendulum — The observed oscillation of engineering orgs between monolith and microservices architectures as they grow, stall, or reorganize — with neither pole being a destination.…
• Monorepo — Monorepo = a single shared version-control repository holding many services, libraries, tools, and other artefacts that would otherwise live in separate repos.…
• Monte Carlo simulation under uncertainty — Monte Carlo simulation evaluates a decision or policy by drawing many random samples from the underlying uncertainty distributions and averaging the outcome across samples.…
• Mountable tiered-storage topic — A mountable tiered-storage topic is a Kafka-API topic whose data already lives in an object store via tiered storage, and which can be safely unmounted from a cluster (hidden…
• MTU IPv6 floor — QUIC, as standardised by the IETF in RFC 9000, does not support network MTUs below 1280 bytes. This matches the IPv6 minimum MTU (RFC 8200),…
• Multi-account abuse — Multi-account abuse is the fraud class where a single fraudulent actor creates several accounts to either:
• Multi-card embedding sharding — Multi-card embedding sharding is the serving-side mechanism that splits a recommendation model's embedding tables into segments distributed across multiple GPUs when the combined…
• Multi-cloud LLM serving — Multi-cloud LLM serving is the architectural posture of running production LLM-powered features against managed model-serving endpoints from two or more independent cloud…
• Multi-dimensional clustering — Multi-dimensional clustering is the table-layout technique where data files are organised to support simultaneous skipping on multiple columns…
• Multi-document ACID transactions — A multi-document ACID transaction bundles multiple document writes — potentially across multiple collections, and in MongoDB's 2019+ versions multiple shards…
• Multi-GPU serving (LLM) — Multi-GPU serving is the LLM-inference regime in which a single model instance spans multiple GPUs because the model's weights + working-set memory exceed a single GPU's VRAM.…
• Multi-hop relationship materialization — Multi-hop relationship materialization is the discipline of walking N-step paths in a graph and writing back direct N=1 edges as new facts,…
• Multi-lane encoding pipeline — A multi-lane encoding pipeline produces multiple video encodings (lanes) from a single source in one coordinated pass. Each lane can differ in resolution, codec, framerate,…
• Multi-language materialized view — A multi-language materialized view is an MV whose definition can be authored in more than one language — typically SQL plus a general-purpose programming language (Python…
• Multi-layer restore validation pipeline — A multi-layer restore validation pipeline is a recovery-time process that combines multiple independent validation checks to prove that a restored backup is safe to use,…
• Multi-LLM sub-agent routing — Multi-LLM sub-agent routing is an agent architecture in which different sub-agents inside a single agent system use different LLMs,…
• Multi-modal attribute extraction — Multi-modal attribute extraction is the pattern of using a vision-language model (VLM) — one that natively takes both image and text inputs…
• Multi-path datacenter transport — A class of non-TCP transport protocols specifically designed to exploit the multiple parallel paths available through modern datacenter fabrics,…
• Multi-region stretch cluster — A multi-region stretch cluster is a deployment topology in which a single streaming broker cluster (or database cluster) spans two or more geographic regions,…
• Multi-shard schema sync — Multi-shard schema sync is the problem of keeping schema identical across every shard of a horizontally- sharded database while a schema change is being rolled out…
• Multi-source topology fusion — Multi-source topology fusion is the architectural choice to build a service dependency graph from multiple independent capture substrates,…
• Multi-step LLM extraction — Multi-step LLM extraction is the discipline of decomposing an information-extraction task into a sequence of narrow LLM invocations,…
• Multi-task learning — Multi-task learning (MTL) is the neural-network training paradigm where a single model with shared parameters is trained jointly on multiple related tasks,…
• Multi-task multi-label (MTML) ranking — MTML ranking is the class of recommendation / search ranker architectures that share a candidate + context representation across a set of tasks (distinct prediction targets…
• Multi-task retrieval scoring — Multi-task retrieval scoring is the practice — first canonicalised on the wiki via Meta's 2026-05-26 SilverTorch post (Source:…
• Multi-tenant graph platform — A multi-tenant graph platform is a shared infrastructure where multiple graph use cases (tenants) run on the same engine with namespace isolation,…
• Multi-tenant LLM capacity allocation (VM-equivalent guarantees) — Multi-tenant LLM capacity allocation is the architectural problem of providing predictable, contractable capacity to individual customers on a shared LLM-serving substrate…
• Multi-Way Join Operator (Flink) — A multi-way join operator in Flink is a single keyed operator that joins N streams at once, holding one shared state per key,…
• Multi-window multi-burn-rate (MWMBR) alerting — Multi-Window Multi-Burn-Rate (MWMBR) alerting is the Google-SRE-Workbook alerting strategy in which an SLO-backed alert fires only when error-budget burn rate exceeds a threshold…
• Multilingual LLM evaluation — Multilingual LLM evaluation is the discipline of measuring LLM behaviour across multiple languages — both within-language capability (does the model perform well in language X?)…
• Multimodal annotation intersection — Multimodal annotation intersection is the operation of fusing annotations produced by independent models of different modalities (vision, audio,…
• Multimodal CPU bottleneck (image preprocessing) — A multimodal CPU bottleneck is a class of LLM-serving performance failure where the CPU-side preprocessing of multimodal inputs (image resize + normalisation,…
• Multimodal Document Understanding — Multimodal Document Understanding is the architectural choice to treat scanned-document pages as images interpreted by a vision-language model rather than as text to be extracted…
• Multiprocessor fraud signal export — Multiprocessor fraud signal export is the architectural shape where a fraud-detection network sells predictive signals as a B2B API for transactions it does not itself process.…
• Muon optimizer — Muon is a neural-network optimiser popularised in 2024 (Keller Jordan et al., jeremyjordan.me / blog posts / arXiv 2502.16982) designed specifically for hidden-layer matrices…
• Mutation history (commit) — Mutation history is VCS metadata that tracks how a commit has evolved over time — every amend, rebase, fold, split, or other mutation produces a new commit,…
• Mutual TLS (mTLS) — Mutual TLS (mTLS) is TLS with two-way peer authentication: both ends of the connection present certificates, and each verifies the other's identity against a trust anchor,…
• MVCC horizon — The MVCC horizon (also "xmin horizon" or "vacuum horizon") is the transaction-id cutoff below which Postgres knows no active transaction can still see old row versions…
• MySQL access type (EXPLAIN type column) — The type column in MySQL EXPLAIN output describes how MySQL will find the rows it needs in a table — the access method, not the join algorithm.…
• MySQL ALGORITHM=INPLACE — ALGORITHM=INPLACE is a MySQL ALTER TABLE execution strategy where InnoDB rebuilds the table in place inside the MySQL server process (without an external shadow-table tool) while…
• MySQL caching_sha2_password authentication — cachingsha2password is the default authentication plugin in MySQL 8.0+, replacing the long-standing mysqlnativepassword plugin that was the default on MySQL 5.7 and earlier.…
• MySQL case-insensitive identifiers — MySQL table and column names are case-insensitive. Postgres identifiers are case-sensitive when quoted in double quotes, and folded to lowercase when unquoted.…
• MySQL CONNECTION_ID vs thread_id — MySQL exposes two distinct integer identifiers for the same running session: the user-facing CONNECTIONID() (aka PROCESSLISTID) and the instrumentation-layer…
• MySQL connection termination at edge — MySQL connection termination at edge is the architectural pattern of fully terminating the TCP connection, TLS handshake,…
• MySQL EXPLAIN ANALYZE — EXPLAIN ANALYZE is the MySQL 8.0.18+ extension that runs the query and reports the actual per-iterator runtime cost alongside the optimiser's estimates.…
• MySQL EXPLAIN — EXPLAIN is the MySQL keyword that asks the query optimizer to describe the execution plan it would choose for a query — without running the query.…
• MySQL index hint — A MySQL index hint is a SQL-level directive that instructs the query optimiser which index to use, or not use, or must use — overriding the planner's own choice.…
• MySQL index statistics — MySQL index statistics are the sampled cardinality values MySQL stores per index to inform the query optimiser's index-selection decisions.…
• MySQL invisible column — A MySQL invisible column is a column that physically exists in a table and participates in explicit-column queries (SELECT colname, INSERT INTO t (colname) VALUES ...),…
• MySQL invisible index — A MySQL invisible index is a secondary index that physically exists and is maintained on every write, but is hidden from the query optimiser — so no query plan will use it.…
• MySQL JSON data type — The JSON data type in MySQL is an engine-level primitive for storing whole JSON documents in a single column with (a) automatic content validation on insert and update,…
• MySQL JSON extraction operator — MySQL exposes two shorthand operators for pulling a value out of a JSON column at a given JSONPath:
• MySQL memory instrumentation — MySQL memory instrumentation is the subset of performanceschema that measures how much memory is allocated + what it's for at server-internal granularity.…
• MySQL partitioning — MySQL partitioning is the native MySQL feature that splits an individual table into several logical units — called partitions — stored separately on disk on the same server.…
• MySQL performance_schema — performanceschema is MySQL's built-in instrumentation subsystem — a database containing ~113 tables that expose server-internal telemetry: per-statement timing,…
• MySQL reserved-words upgrade break — The class of MySQL upgrade breakage where a new major version promotes previously-legal identifiers to reserved words, making existing schemas unparseable on the new version until…
• MySQL semi-sync replication — MySQL semi-synchronous replication (semi-sync) is a plugin mechanism layered on top of MySQL's native asynchronous replication that blocks a commit on the primary until…
• MySQL semi-sync split-brain — A MySQL semi-sync split-brain is a specific production hazard in which a MySQL primary using semi-synchronous replication completes writes that were never durable…
• MySQL Snapshot to S3 Data Lake — MySQL Snapshot to S3 Data Lake is the pattern of regularly (typically daily) copying the full state of MySQL operational tables into immutable Parquet / Avro / JSON files on S3,…
• MySQL spatial binary (WKB) vs Postgres Point — MySQL and Postgres both declare a POINT column type. The declaration is identical, but the wire representation at read time is not.…
• MySQL sql_mode — sqlmode is a MySQL server variable that controls SQL-dialect strictness — which queries the server silently tolerates vs rejects, which implicit type coercions are allowed,…
• MySQL sys schema — The sys schema is a curated set of views + stored procedures over performanceschema that present server telemetry in human-readable form.…
• MySQL TEMPORARY keyword in DROP TABLE — MySQL's DROP TABLE supports a TEMPORARY keyword that scopes the drop to a temporary table only. Postgres does not. The absence makes a regular-table drop under the same name a real…
• MySQL transaction isolation levels — MySQL transaction isolation levels are the four modes that can be set on a MySQL session (or globally, or at server startup) to control how concurrent transactions interact…
• MySQL undo log — The undo log is InnoDB's mechanism for reconstructing previous versions of a row on demand. Unlike Postgres — which keeps multiple row versions on the heap and garbages them later…
• MySQL VARCHAR length requirement — MySQL VARCHAR declarations require an explicit maximum length. Postgres allows VARCHAR without a length, behaving like an unbounded text column.…
• MySQL version upgrade — A MySQL version upgrade is the operational primitive of moving a running MySQL database from version X.Y.Z to a later version.…
• Named contexts for multi-stack — Named contexts (kubectl-style) are a pattern for letting a single CLI address multiple target environments (stacks, clusters, accounts,…
• Named graph — A named graph is an RDF construct that attaches an identifier (a URI) to a set of triples — letting the knowledge graph partition triples into addressable subgraphs instead…
• Namespace organization — Namespace organization in a data warehouse uses namespaces to define clear placement for every table, preventing confusion about ownership and scope as a multi-product system…
• NAND flash page/block erasure semantics — NAND flash is organised in a three-level hierarchy with asymmetric unit sizes for read, write, and erase:
• NanoID collision retry — NanoID collision retry is a bounded application-side retry loop that regenerates a random identifier on collision — before INSERT, if checked by exists?, or after INSERT,…
• NanoID — A NanoID is a URL-safe random string identifier (Andrey Sitnik, 2017) designed as a compact, collision- resistant alternative to UUIDs.…
• Narrative coherence as hallucination filter — Narrative-coherence-as-hallucination-filter is the architectural technique of running a second pruning pass over already-credibility-scored findings in which each finding…
• NarrowBand vs WideBand audio — Audio codecs operate in one of several band modes defined by the frequency range they preserve. The common modes (per ITU-T G.722 / G.711 / codec literature):
• National grid collapse — A national grid collapse is a cascading failure of a country's interconnected electricity-transmission system such that power generation and delivery fail across the country (or…
• Natural-language infrastructure provisioning — Natural-language infrastructure provisioning is the UX posture where the primary human-to-infrastructure interface is a free-form conversation with an LLM — not a YAML manifest,…
• Navigation mix — Navigation mix is the measured distribution of navigation classes reaching a given route in a web application — typically taxonomised as hard / partial / soft with cost shapes…
• NDCG (Normalized Discounted Cumulative Gain) — Normalized Discounted Cumulative Gain (NDCG) is a standard information-retrieval metric for ranked list quality. It measures how well a ranked retrieval result matches the ideal…
• NDJSON streaming format — NDJSON (Newline-Delimited JSON; also called JSON Lines or jsonl) is a wire format where each record is a single self- contained JSON object on its own line, separated by \n.…
• Near-atomic schema deployment — Near-atomic schema deployment is the property that a set of schema changes grouped into one deployment unit (a PlanetScale deploy-request, a migration branch) cuts over together,…
• Needle-in-haystack log query — A needle-in-haystack log query is a log search for a single highly-unique value — a UUID, a request ID, a trace ID, a job ID, a user ID — across a large corpus of logs.…
• Negative caching — Negative caching is caching empty / no-result / absence-of-data responses so the cache doesn't re-query the backend every time for a legitimately-empty answer.…
• Negative transfer — Negative transfer is the failure mode of transfer learning / domain-adaptive learning in which transferring knowledge from a source domain degrades performance on the target…
• Negative Trust Anchor — A Negative Trust Anchor (NTA), defined in RFC 7646, is an explicit exception on a DNSSEC-validating resolver that tells it to treat a specific zone as if it were unsigned,…
• NER-clustered Query Sampling — NER-clustered query sampling is the query-sampling discipline of grouping production search queries by their NER-extracted attribute set (category / brand / colour / size / season…
• NER-tag Parity Across Languages — NER-tag parity across languages is the operational realisation of the translated-query-parity invariant: run the same NER engine against both the source-language query and its…
• Nested document indexing — Nested document indexing is the Elasticsearch document structure where a single top-level document carries a field typed as nested whose values are an array of independently-…
• Nested-loop join — A nested-loop join (NL join) is the join algorithm that iterates over the left-hand side (LHS) rows one at a time and, for each LHS row,…
• Nested-sort path requirement — When sorting Elasticsearch search results on a nested field (a field inside an Elasticsearch nested type), the sort clause must include a nested.path sub-clause naming the path…
• Network-attached storage latency penalty — Network-attached storage latency penalty is the extra round-trip time a storage IO pays when the backing drive sits across a network hop from the compute instance,…
• Network boot interface — A network boot interface is a firmware-level mechanism that lets a server boot its operating system over the network instead of from local storage.…
• Network-bound vs compute-bound — A system is network-bound when its scaling-limiting resource is network bandwidth (or packet rate) — adding more CPU / GPU does not increase throughput because bytes-per-second…
• Network driver reset — A network driver reset (or "device reset") is a self-healing path in a Linux network driver — the driver concludes the hardware or its own kernel threads have stopped making…
• Network-effect fraud detection — Network-effect fraud detection is the architectural thesis that a fraud-detection network's value grows super-linearly with the number of payment methods covered,…
• Network fabric disaggregation — Network fabric disaggregation is the architectural stance of splitting a vertically-integrated network fabric — switch ASIC + switch hardware + NOS + control plane + endpoint NIC…
• Network intermediary resolution — Network intermediary resolution is the operation of collapsing multi-hop network flows that pass through intermediaries — load balancers, NAT gateways, API gateways,…
• Network round-trip cost — The round-trip-time (RTT) floor between an application process and a remote database or RPC service is the unit cost that dominates batch-job throughput whenever a loop does one…
• Neurosymbolic AI — Neurosymbolic AI is the composition of neural methods (LLMs, transformers, RL) with symbolic methods (mechanical theorem provers, SAT/SMT solvers, formal reasoners).…
• Neutral vs non-neutral parameter — In the context of URL normalisation, a query parameter is:
• NewReno congestion control — NewReno is a loss-based TCP congestion-control algorithm standardised in RFC 6582 (2012) — a refinement of Reno (originally 1990) dating back to 1999.…
• Nil-index Lua bug — A Lua error raised when code attempts to index a field on a value that turns out to be nil:
• NMSE (Normalized Mean Squared Error) — Normalized Mean Squared Error (NMSE) is the graded-scale judge-vs-human agreement metric used by Dropbox Dash's relevance judge (Source:…
• No container image (Sprite design) — A VM-platform design decision: do not let users supply a container image. Every instance boots from a single platform- standard image;…
• No distributed consensus — "No distributed consensus" is a deliberate design choice: build the state-distribution system without relying on a consensus protocol (Raft, Paxos, Multi-Paxos, Zab,…
• No message-history carry-forward — No-message-history-carry-forward is the architectural discipline of not carrying raw accumulated LLM message history between agent invocations in a long-running multi- agent loop.…
• Nodeless Kubernetes — Nodeless Kubernetes is the architectural stance of running a Kubernetes API surface without any real Node objects or worker VMs under the control of the cluster operator.…
• noindex meta tag — The noindex meta tag is an HTML directive, typically in the page , that tells search engines not to include the page in their index:
• Noise injection in evaluation — Noise injection in evaluation is the counter-intuitive discipline of making the simulated environment an agent is evaluated in messier, not cleaner — by including signals,…
• Noise Protocol Framework — The Noise Protocol Framework is Trevor Perrin's specification framework for cryptographic handshake protocols built out of Diffie-Hellman, AEAD ciphers, and hash functions.…
• Noisy neighbor — Noisy neighbor names the multi-tenant failure mode where one tenant's workload perturbs another tenant's latency/throughput — through shared queues, shared media, shared CPU,…
• Non-deterministic MV maintenance — Non-deterministic MV maintenance is the IVM problem of keeping a materialized view correctly up to date when the view definition includes non-deterministic functions…
• Non-idempotent endpoint parallel-run constraint — The non-idempotent endpoint parallel-run constraint is the applicability limit of the parallel run pattern: endpoints with side effects that cannot be double-executed safely (POSTs…
• Non-idempotent operations — A non-idempotent operation is one whose effect changes with each execution: retrying a successful-but-ambiguously-failed call risks duplicate side effects (double-charged users,…
• Non-inferiority test — A non-inferiority test is a statistical hypothesis test designed to show that a treatment is not meaningfully worse than the control, within a pre-specified margin.…
• Non-Maximum Suppression (NMS) — Non-Maximum Suppression (NMS) is the classical object- detection post-processing technique that de-duplicates overlapping bounding-box detections by keeping the highest-confidence…
• Non-repeatable read — A non-repeatable read is a read-anomaly class that occurs when a transaction issues the same SELECT twice and gets different values for the same row between the two reads…
• Non-revertible schema change — A non-revertible schema change is a schema-change path whose execution does not build the substrate required for an automated, no-data-loss revert.…
• Non-targetability — Non-targetability is the security property that an attacker cannot target a specific user for compromise without attempting to compromise the entire system.…
• Non-uniform LLM request cost — Non-uniform LLM request cost is the structural property of LLM inference where request cost varies by 1-3 orders of magnitude across the natural request distribution,…
• NoSQL database — A NoSQL database is a non-relational datastore — it does not require data to be laid out in fixed-schema tables joined via foreign keys and accessed through SQL.…
• Notebook Experimentation Platform — A notebook experimentation platform is a hosted, multi-user, pre-wired notebook environment offered by an internal ML or data-platform team as the default first-contact surface…
• Notebook format migration — Notebook format migration is the act of moving an interactive analytics notebook (and its embedded code, visualisations, widgets,…
• Notification-of-change event — A notification-of-change event is an event whose payload contains only the fact that something changed — typically an entity identifier and an event type — but not the new state.…
• Nullable-column backfill amplification — A nullable-column backfill amplification is a specific operational hazard in CRDT-backed distributed databases where adding a nullable column to a table forces the CRDT engine…
• numHistoryShards immutability — numHistoryShards is the shard count of Temporal's History subsystem — the value that determines how many partitions Temporal will hash workflows across.…
• NUMA awareness — NUMA (Non-Uniform Memory Access) is the dominant memory architecture on multi-socket x86 servers: each CPU socket has its own memory controller + attached DRAM.…
• Nvidia driver happy path — The Nvidia driver "happy path" is the shape of host configuration that Nvidia's proprietary driver stack is engineered to support turnkey: a Linux host running either (a)…
• OAuth JWT short-lived credential — An OAuth JWT short-lived credential is an OAuth 2.0-issued JSON Web Token with a bounded TTL (minutes to hours, not days to forever),…
• OAuth Protected Resource Metadata — OAuth Protected Resource Metadata (RFC 9728) is a well-known document published at /.well-known/oauth-protected-resource that tells a client where to find the authorization server…
• Object-capability RPC — Object-capability RPC is an RPC model in which references to remote objects and functions are first-class values on the wire.…
• Object storage as disk root — A storage-architecture decision: the authoritative durability tier for a VM's disk is an S3-compatible object store, not local NVMe.…
• Object-tree document model — An object-tree document model represents a document as a rooted tree of objects, each with a stable ObjectID and a bag of (Property, Value) entries.…
• Objective abstraction — An Objective is an enumerated identifier for a business use case in a model-serving platform — a name like ContinueWatchingRanking or PaymentFraudDetection that clients use…
• Oblivious HTTP (OHTTP) — Oblivious HTTP (OHTTP; RFC 9458) is a protocol that allows a client to send an HTTP request to a gateway such that no single party learns both the client identity (IP address)…
• Observability as code — Observability as code is the posture of treating every observability resource — dashboards, alert rules, SLOs, synthetic checks, recording rules,…
• Observability before migration — Close the observability gap on the new path before the migration proceeds. Orgs that switch transports / protocols / platforms without first extending their telemetry to cover…
• Observability SDK wrapper — An observability SDK wrapper is a thin, internally-owned library that sits on top of a standards-based observability SDK (e.g.…
• Observability stack partial dependency — An observability stack partial dependency is the failure-mode property of an observability architecture that splits responsibilities across providers…
• Observability traffic volume asymmetry — Observability traffic volume asymmetry is the architectural property that at scale, an org's observability telemetry traffic (metrics scrapes, remote-writes, logs,…
• Observability — The function of providing visibility into application performance and reliability via metrics, logs, and traces. The core operational quality it serves: lowering MTTD (mean time…
• ODR violation (One-Definition Rule) — The One-Definition Rule (ODR) is a C++ language rule that requires every non-inline external symbol (function, variable, class type,…
• Offense / defense performance engineering — Offense / defense performance engineering is Meta's operational framing of hyperscale capacity efficiency as a two-sided problem that requires investment in both halves to keep…
• Offline compute / online lookup — Offline compute / online lookup is an architectural split in which an expensive analysis runs asynchronously on a batch cadence, produces a small configuration artefact,…
• Offline data warehouse as translation layer — The offline (analytics-oriented) data warehouse acts as a crucial translation layer between upstream online/OLTP systems…
• Offline query tuning loop — The offline query tuning loop is the workflow by which a query's execution plan is improved outside the query optimizer hot path, typically:
• Offset-commit cost — Offset-commit cost names the often-overlooked broker-side tax that consumer groups pay every time they commit their progress.…
• offset.mismatch.strategy (Debezium) — offset.mismatch.strategy is a Debezium Postgres connector configuration property introduced in Debezium 3.4.0.Final (2025-12-16) that controls what the connector does on startup…
• Offset pagination cost — The runtime cost of SELECT … ORDER BY key LIMIT N OFFSET M in most relational databases is proportional to M + N, not N. OFFSET is implemented as fetch the first M+N rows in order…
• Offset planning — Offset planning is a named pipeline stage in the Vitess query planner introduced as part of the "new model" rewrite (canonicalised in Andrés Taylor's 2023-06-01 post).…
• Offset-preserving replication — Offset-preserving replication is cross-cluster replication where the destination cluster holds the same per-partition offsets as the source cluster.…
• Offsite conversion sparsity — Offsite conversion sparsity is the structural training-data problem faced by ads-ML systems optimising for conversion actions (purchase, checkout, add-to-cart, sign-up,…
• OIDC federation for cloud access — OIDC federation for cloud access is the architectural pattern where a workload running in platform A (e.g. Fly.io, GitHub Actions,…
• OIDC identity federation — OIDC identity federation is the pattern of using an OpenID Connect identity provider (IdP) to issue short-lived, signed JWTs that downstream services verify and exchange for their…
• OLTP vs OLAP — OLTP (Online Transaction Processing) and OLAP (Online Analytical Processing) are the two big workload archetypes databases are tuned for.…
• OMP_NUM_THREADS container misconfiguration — The OMPNUMTHREADS container misconfiguration is a multi-tenant Kubernetes / container-runtime pathology where libraries that read the OMPNUMTHREADS environment variable (PyTorch,…
• On-call health metric — On-call health is a first-class, measurable KPI for an engineering team's on-call program — not a soft concern. The standard definition composes paging rate (how many pages per…
• On-call rotation — An on-call rotation is the schedule that assigns a small set of engineers to be reachable for production incidents during a defined window — typically 24×7,…
• On-demand feature compute — On-demand feature compute is the feature-store shape where feature values are not precomputed up-front; instead, the store responds to a request by computing the feature dependency…
• On-Device ML Inference — On-device ML inference is the running of ML inference on end-user hardware — smartphones, laptops, browsers, embedded devices, NPUs — rather than on cloud servers.…
• On-policy RL vs SFT signal shape — The signal shape of an LLM post-training method — whether the learning signal is dense and immediate (per-token, per-batch,…
• On-the-fly security principal rewrite — On-the-fly security principal rewrite is the design primitive of translating a request's authenticated identity into a different principal at an ingress service,…
• One definition, many runtimes — One definition, many runtimes is a platform organising principle: "Define a signal or event type once, then instantiate it consistently across multiple runtimes." (Source:…
• One-to-one agent instance — The one-to-one agent instance is the structural observation that an AI agent, unlike a traditional request handler or web service,…
• Online context summarisation — Online context summarisation is the pattern of compacting state round-by-round as an investigation progresses — as opposed to offline summarisation (done retrospectively after…
• Online database import — Online database import is the capability to move an existing production database from one vendor or host to another without downtime…
• Online DDL — Online DDL is the family of techniques for applying schema changes (ALTER TABLE, CREATE INDEX, column adds, type changes,…
• Online-offline discrepancy — Online-offline discrepancy (O/O discrepancy) is the named production hazard where a change to an ML model shows clear offline wins on standard evaluation metrics (loss,…
• Online scheduling — Online scheduling is the problem class where "jobs arrive dynamically and the scheduler must make immediate, irrevocable decisions without knowing what jobs will arrive next"…
• Online-stateful token — An online-stateful token requires a database lookup at the token authority on every verification — in contrast to a signed,…
• Online vs offline feature store — An online feature store and an offline feature store are the two complementary faces of the same feature repository, tuned for different access patterns:
• Only-to-Customer (OTC) attribute (RFC 9234) — OTC is a BGP path attribute defined in RFC 9234 that, together with a new BGP Role capability negotiated at session setup,…
• Opaque attribute code translation layer — An opaque attribute code translation layer is a bidirectional shim that translates internal identifier codes into human-readable natural language on the way into an LLM,…
• Opaque output-format fencing — Opaque output-format fencing is the prompt-engineering primitive of instructing the LLM to wrap its entire response in a pair of opaque delimiter tags…
• Open File Format — An open file format is a non-proprietary, language-agnostic, object-level data representation for tabular data on storage. The canonical pair in the 2020s:
• Open media standards — Open media standards in the film + TV production domain are publicly specified, cross-vendor file-interchange formats that encode colour, framing, media hashes,…
• Open Table Format — An open table format (OTF) is a metadata layer over columnar data files on object storage that adds table semantics — atomic row-level updates, schema evolution,…
• OpenTelemetry collector — receiver / processor / exporter pipeline — The OpenTelemetry Collector is OTel's vendor-neutral intermediary between data sources (applications, SDKs, infra agents,…
• Operation-based SLO — An Operation-Based SLO is a Service Level Objective defined per business operation / user journey (e.g., "checkout success rate ≥ 99.9%",…
• Operational ↔ analytical governance unification — Operational ↔ analytical governance unification is the architectural property that the operational tier (the OLTP database serving the application) and the analytical tier (the…
• Operational efficiency vs risk isolation trade-off — The operational efficiency vs risk isolation trade-off is the canonical design tension in multi-account enterprise architecture: more isolation boundaries buy stronger blast-radius…
• Operational relational schema paradigm — The operational relational schema paradigm is Shlomi Noach's 2022 framework of ten tenets an operationally-modern relational database must satisfy for schema change to feel like…
• Operational Transform (OT) — Operational Transform (OT) is the dominant real-time-collaboration algorithm of the Google-Docs / Etherpad / pre-2015 era.…
• Operator reconcile abort on spec change — The reconcile-abort-on-spec-change contract is the behavioural property of a well-written Kubernetes operator that, when a CRD instance's desired state changes mid-reconcile,…
• Operator-to-event ratio — The operator-to-event ratio is the scaling primitive for live broadcast operations: how many concurrent events a single human operator can responsibly oversee in a given role.…
• Opt-in marker interface — An opt-in marker interface is an interface (or protocol, or trait, depending on language) whose sole purpose is for an implementor to declare opt-in participation in some…
• Opt-in PII redaction per session — A data-platform governance shape in which PII columns are redacted by default before query results ever reach the user, and legitimate access flips a session-level bit…
• Optimistic locking — Optimistic locking is a concurrency-control discipline in which a writer reads a row (including a version column or equivalent monotonic marker),…
• Optimizer index-expression rewrite — Optimizer index-expression rewrite is the query-planner capability of recognising that a WHERE predicate expression matches the expression underneath an indexed generated column…
• Optimizer statistics as skipping substrate — The architectural framing that table statistics are not just plan-quality input for a cost-based optimiser — they are the substrate that makes data skipping possible.…
• Oracle LogMiner CDC — Oracle LogMiner CDC is the per-engine change-data-capture mechanism that consumes Oracle LogMiner — the Oracle Enterprise Edition redo-log-mining utility…
• Organizational unit (AWS OU) — An organizational unit (OU) is the nested-grouping primitive inside an AWS Organization: AWS accounts are grouped into OUs by some shared axis (business unit, environment,…
• Origin-bound credential — An origin-bound credential is an authentication credential that is cryptographically tied to the web origin — the (scheme, host, port) tuple — it was registered against.…
• Orphan lookup row — An orphan lookup row is a row in a Vindex lookup table whose referenced user row no longer exists — typically because the Consistent Lookup Vindex's Post connection failed…
• OS-library vulnerability ungovernable — An OS-library vulnerability ungovernable is a bug in a library provided by the operating system — not bundled with the app — which the app developer cannot patch directly.…
• OTLP Profiles signal — The OTLP Profiles signal is the OpenTelemetry project's wire- format and semantic-convention standard for continuous profiling data, reaching alpha status in early 2026.…
• Out-of-sequence CDC event handling — Out-of-sequence CDC event handling is the class of correctness problems that arise when CDC events arrive at the consumer in a different order than the logical order they occurred…
• Outbound proxy credential injection — An agent-egress security pattern in which secrets are never made available inside the agent's sandbox. Instead, every outbound request from the sandbox passes through…
• Over-partitioning — Over-partitioning is the structural failure mode where a Hive-style partitioned table is laid out at finer granularity than the workload requires, producing many small files,…
• Overall Evaluation Criterion (OEC) — The Overall Evaluation Criterion (OEC) is the single composite metric an A/B test (or family of tests) is decided on — the metric that,…
• Overfitting to historical data — Overfitting to historical data is the failure mode of any back-testing / historical-replay methodology where optimisation against past data produces candidates that win on history…
• Overlapping rollouts — Overlapping rollouts is the architectural acceptance that in a sufficiently large fleet, multiple concurrent upgrade campaigns — touching different components,…
• Oversampling metric interval — For a metric used as a threshold-based control-loop input, sample at 2–5× the rate of the threshold range you care about.
• Ownership (Amazon's organizational primitive) — Ownership, at Amazon, is the organizational primitive that says a single person or team is end-to-end accountable for a service or outcome: the API contracts, the durability,…
• Packet sniffing as event source — Packet sniffing as event source is the move of manufacturing a control-plane event by observing the data plane — when the primitive you care about exposes traffic but not…
• Padding removal (variable-length inference) — Padding removal (also called variable-length processing) is the inference-engine technique of serving a batch of variable-length transformer inputs as a single concatenated…
• Pagecache for messaging — Pagecache is the OS kernel's in-memory cache of disk pages — any file read (first time) pulls the page into pagecache, so subsequent reads of the same page are served from RAM…
• panic=abort — panic=abort is Rust's abort-on-panic strategy: when panic!() fires, the process terminates immediately with no stack unwinding and no destructors run.…
• panic=unwind — panic=unwind is Rust's unwinding panic strategy: when panic!() fires (or .unwrap() on an Err, or an assertion fails), the runtime walks the stack backwards,…
• Parallel broker replication tasks — Parallel broker replication tasks is the broker-internal cross-cluster replication property where every broker in the shadow / destination cluster runs its own replication tasks…
• Parallel cross-and-deep network — A parallel cross-and-deep network is a feature-crossing neural architecture where an explicit-feature-cross tower (typically DCNv2) and a deep feed-forward tower (MLP) operate…
• Parallel recovery stages — Parallel recovery stages is the recovery-workflow design principle that independent stages run concurrently to compress the critical path of a cyber-event recovery.…
• Parallel-run request doubling — Parallel-run request doubling is the operational cost envelope of running a parallel run: every incoming request is processed twice…
• Parallel snapshot (intra-table CDC) — Parallel snapshot is a CDC-connector optimisation that splits a single large table or collection into chunks and reads those chunks concurrently during the snapshot phase…
• Parallel thinking (trajectory sampling) — Parallel thinking is an agent-design technique where the agent samples multiple independent trajectories of reasoning over the same query and aggregates findings across them…
• Parameter binding — A parameter binding is the edge in a parameter system that says "this specific layer property reads its value from this specific parameter." The layer property stops storing…
• Parameter system — A parameter system defines values that can be set once and applied across many consumers. Change the value in one place → all consumers update.…
• Parameterized workflow — A parameterized workflow is a workflow whose structure (number of steps, which steps run, values passed between steps) is initialised step-by-step at runtime based on user-defined…
• Parser differential — A parser differential is a security-relevant disagreement between two implementations of the same parsing contract: given the same input bytes,…
• Partial failure — Partial failure is the state of a distributed system where a component is neither fully up nor fully down — it is running but delivering degraded correctness, throughput,…
• Partial network partition — A partial network partition is a connectivity failure in which a node can reach some of its network peers but not others,…
• Partial restoration pattern — The partial restoration pattern is the recurring shape of Internet-connectivity recovery following a government-directed shutdown: connectivity comes back in phases rather than…
• Partition access pattern — A partition access pattern is the signature shape that emerges when you plot the partition values accessed against access-event time for a single analytics table…
• Partition colocation (cross-topic) — Partition colocation (cross-topic) is the property that same-index partitions of multiple topics are assigned to the same processing instance / task.…
• Partition projection (Glue / Athena) — Partition projection is the Athena / Glue partitioning model where the set of partition values is declared in the table properties rather than registered as managed partitions…
• Partition pruning — Partition pruning is the query-optimizer technique whereby the planner uses a query's WHERE predicates, combined with a partitioned table's partitioning function metadata,…
• Partition quality marking — Partition quality marking is the practice of annotating a data partition's metadata with a quality flag (good / bad / unknown) so that downstream consumers + the ingestion system…
• Partition skew (data skew) — Partition skew — also known as data skew — is the streaming-broker failure mode where records are distributed unevenly across a topic's partitions,…
• Partition strategy — Partition strategy is the umbrella term for "how a sharded database decides which rows are stored together and on which server". Justin Gage's canonical naming (Source: ):
• Partitioning vs sharding — Partitioning and sharding are often confused because both split a table into smaller pieces. The distinction is on where the pieces live and who is responsible for splitting them:
• Partner portal interactive planning — Partner portal interactive planning is a B2B UX shape where partners / merchants / suppliers interact with a vendor-hosted portal that exposes data + decision recommendations +…
• pass@k — pass@k asks: for a given scenario, over k independent attempts, does the agent succeed on at least one of them?
• Pass-through write to object store — Pass-through write to object store is the storage-path shape where a streaming broker receives a message, writes the message payload directly to object storage (S3 / GCS / ADLS)…
• Passive vs active client signals — Cloudflare's 2026-04-21 post taxonomizes the signals an origin server observes from a client into three categories: passive client signals, active client signals,…
• Passkey authentication — Passkey authentication is a passwordless authentication primitive built on the WebAuthn / FIDO2 standards. A passkey is a public-private keypair generated and stored on the user's…
• Patch lag — Patch lag is the time between a security fix being released and the user actually running the fixed software. For server-side systems,…
• Path enumeration attack — A path enumeration attack is the reconnaissance technique of systematically requesting URLs against a target to discover hidden, private,…
• Path pinning (RoCE routing) — Path pinning is a routing scheme where specific packets are deterministically assigned to specific parallel paths through a multi-path fabric — the opposite of hash-based ECMP.…
• Path validation (BGP) — Path validation is the class of mechanisms that validate not just the origin of a BGP route (what ROV does) but the entire AS path against the relationships it implicitly claims.
• Pay-as-you-go abuse — Pay-as-you-go abuse (or consumption-billing abuse / post-paid-nonpayment abuse) is the fraud class where customers exploit the structure of consumption-based pricing…
• Payment token over credit card sharing — The principle that raw payment details (credit card numbers, CVVs, full PANs) never reach the agent or the external provider being provisioned;…
• Peer-peer relationship (BGP) — A peer-peer relationship is one of the two primary pairwise business relationships between ASes: two networks exchange traffic settlement-free — neither pays the other…
• Peer sampling service — The peer sampling service (PSS) is the abstraction that hands a gossip implementation its next peer on each round. It hides the question "who do I talk to?" behind a small API,…
• PEG grammar (Parsing Expression Grammar) — A Parsing Expression Grammar (PEG) is a formal grammar class (Ford 2004) defined by ordered choice (A | B means try A first, fall back to B only if A fails),…
• Per-account quotas — Most AWS service limits ("service quotas") are enforced per AWS account. In a shared-account architecture this is a single, easily-monitored number;…
• Per-channel vs per-tensor FP8 scaling — FP8 scaling granularity is the choice of how many FP8-quantised elements share a single floating-point scale factor. Two ends of the spectrum:
• Per-core CPU visibility — Per-core CPU visibility is the discipline of watching CPU utilisation core-by-core rather than as a whole-machine aggregate.…
• Per-hour + per-minute rollup tables — Per-hour + per-minute rollup tables is the pattern of pre-aggregating time-series telemetry at two different time granularities so that queries of different window sizes hit…
• Per-interval tag combination limit — The technique of bounding the number of unique tag combinations emitted per query pattern per aggregation interval, even when every individual tag key is low-cardinality.…
• Per-pattern tag cardinality — The technique of tracking the distinct-value count for a tag key within the scope of a single query pattern (rather than globally across all patterns),…
• Per-pattern time-series — A per-pattern time-series is a stream of aggregated metrics — count, total execution time, rows read, rows returned, rows written,…
• Per-PR ephemeral environment — A per-PR ephemeral environment is an isolated, named environment whose lifecycle is bound to a single pull request: created when the PR is opened,…
• Per-priority AIMD coefficients — Per-priority AIMD coefficients is the technique of running AIMD rate loops for multiple traffic classes using different increase and decrease constants per class.…
• Per-query used-index set — The per-query used-index set is the collection of indexes the storage engine touched during a single query execution. For a query hitting a composite index once it's a singleton…
• Per-request billing — A cloud billing model where the unit of charge is the individual request / invocation, rather than provisioned capacity (vCPU-hours, GB-hours of RAM).…
• Per-tenant dynamic code dispatch — Per-tenant dynamic code dispatch is the runtime shape in which a platform hosts a single, statically-registered entry point,…
• Per-tenant rate limit — A per-tenant rate limit is a rate-limiting budget that a shared service applies independently per tenant rather than globally across all tenants.…
• Per-tenant retention via partitioning key — Per-tenant retention via partitioning key is the multi-tenant-analytics design idiom of extending the partitioning key with a tenant-identifier column so that each tenant's data…
• Per-tenant search instance — A per-tenant search instance is an isolated (storage, index) unit, one per tenant (agent, customer, session, language, region, …), created and destroyed at runtime,…
• Per-tenant workspace isolation — Per-tenant workspace isolation is the multi-tenancy shape where the tenant boundary is an opaque "workspace" inside a third-party SaaS that the platform provisions and operates…
• Per-thread memory profiling — Per-thread memory profiling is the MySQL technique of measuring a running query's memory usage by sampling the thread running that query,…
• Per-topic batch diagnosis — Per-topic batch diagnosis is the discipline of measuring effective batch size disaggregated by topic, not as a cluster-wide average.…
• Per-topic granularity failover — Per-topic granularity failover is the property of a cross- cluster disaster-recovery mechanism where failover can be invoked for an individual topic (promoting its shadow…
• Per-transaction human approval for agent spend — Per-transaction human approval for agent spend is a safety posture in which an AI agent cannot complete a payment without the human user explicitly approving that specific spend…
• Percentage rollout — A percentage rollout serves a variation to X% of the requests that match a rule — so a flag can be ramped 5% → 10% → 50% → 100% of users over time while holding the architecture…
• Percentile-objective optimisation — Percentile-objective optimisation is a risk-aware stochastic optimisation formulation in which the decision θ is chosen to minimise a high percentile of the cost distribution…
• Perceptual conform matching — Perceptual conform matching is the use of computer-vision content-similarity probes — rather than metadata identifiers — to resolve references from an editorial timeline (EDL) back…
• Performance isolation — Performance isolation is the property that one tenant's workload does not observably affect another tenant's latency or throughput…
• Performance per watt — Performance per watt is the ratio of useful work (throughput, operations completed, queries served) to electrical power consumed.…
• Performance prediction — Performance prediction is the problem class of estimating a system's performance metric — throughput, latency, efficiency, resource cost — from a description of its state,…
• Performance regression from mid-upgrade state — A performance regression from mid-upgrade state is a latency / throughput degradation observed during a rolling upgrade while the cluster is in a mixed-version state…
• Performance-variance degradation — Performance-variance degradation is a failure mode where a storage / compute / network substrate delivers some delivered performance level on some fraction of time…
• Permissions DSL — A permissions DSL is a dedicated domain-specific language for expressing authorization rules as data separate from application code. A rule declares at minimum:
• PgBouncer connection chain — The PgBouncer connection chain is the cascade of configuration caps that govern how client connections reach Postgres through PgBouncer.…
• PgBouncer deployment topology — PgBouncer deployment topology is the placement choice for where PgBouncer runs relative to the Postgres primary / replicas.…
• PgBouncer pool sizing formula — The canonical PgBouncer sizing invariant is a single inequality:
• Phantom read — A phantom read is a read-anomaly class that occurs when a transaction issues the same SELECT twice and gets a different number of rows between the two reads…
• Phishing-resistant authentication — Phishing-resistant authentication is authentication whose credentials cannot be replayed against a proxy site because the authenticator cryptographically binds the sign-in…
• Physical resource isolation — Physical resource isolation means running workloads with distinct resource-demand profiles on separate physical machines (or at least separate VMs / containers with hard resource…
• Pickup spot recommendation — Pickup spot recommendation is the ride-sharing / on-demand- logistics problem of choosing where the rider and driver should actually meet when the rider's current GPS position…
• PID controller (feedback control) — A PID controller (proportional–integral–derivative) is a 90-year-old control-theory primitive that drives a process variable toward a setpoint by computing a correction from three…
• Pinned (page-locked) memory — Pinned memory (or page-locked memory) is host-side RAM that the operating system has been told must not be paged out to disk and must not be relocated to another physical address.…
• Pipeline environment — A pipeline environment is a named version of a batch pipeline — a complete set of orchestrator workflow definitions (e.g.…
• Pipeline parallelism — Pipeline parallelism is a multi-GPU model-sharding strategy in which different transformer layers live on different GPUs — GPU 0 holds layers 1-N/G,…
• PK Token — A PK Token is an OpenID Connect ID Token extended to commit to a public key. Introduced by the OpenPubkey project (Linux Foundation, 2023).…
• Placeholder batch (metadata in Raft) — A placeholder batch is a small, metadata-only record replicated through a streaming broker's Raft log whose payload is a pointer (object-storage URL, file offset,…
• Placement-theme cohesion — Placement-theme cohesion is the quality property of a multi-section recommendation page that every section (placement) is intentionally grouped, ordered,…
• Plan cost to wallclock constant k — The k constant is a per-(query-pattern × host) time-varying calibration factor that translates Postgres's dimensionless planner cost estimate into an estimated wall-clock execution…
• PlanetScale hard transaction/query timeouts — PlanetScale enforces two hard timeouts at the vttablet layer that bound every client interaction with the database — 20 seconds for any open transaction and 900 seconds (15…
• Planner phase ordering — Planner phase ordering is the discipline of grouping query-planner rewrite rules into sequential phases, where each phase has its own active rewriter set and runs to a fixed point…
• Platform-asymmetric rollout control — Platform-asymmetric rollout control is the observation that iOS and Android offer materially different control surfaces for staged production rollouts,…
• Platform-specific import resolution — Platform-specific import resolution is the bundler-level mechanism of resolving a single import statement to different files per target platform based on file extension.…
• Platform team bottleneck — The platform team bottleneck is the organisational anti-pattern where a centralised platform / infrastructure / DevOps team becomes the single serialisation point for every…
• Platform team vs application team split — An explicit organisational boundary where *platform engineers own how a capability runs (storage, engine, telemetry, bundle distribution,…
• Playbook expiry date — A playbook expiry date is a per-playbook timestamp that triggers a re-review of the playbook once reached. The expiry-and-renew loop is a forcing function against silent drift…
• Playbook ordering by business impact — Playbook ordering by business impact is the discipline of sorting the set of incident playbooks for a given system from least business impact → most business impact,…
• Playwright Locator auto-wait — Playwright's Locator is the object representing a way to find an element on a page. Unlike a static handle, it re-resolves the selector on each action and builds in auto-waiting:…
• Pluggable persistence layer — A pluggable persistence layer is a storage architecture where each logical tier of a data structure (WAL, data files, manifest/metadata) can be independently backed by a different…
• Pluggable storage backend — A pluggable storage backend is an architecture where the application logic layer (query engine, computation engine) is decoupled from the persistence layer via a well-defined…
• Pluggable validation framework — A pluggable validation framework is a control-plane contract: applications register custom validation hooks that run at the relevant lifecycle events (create, update, delete),…
• Plugin marketplace lock-in — Plugin marketplace lock-in is the structural dynamic where an insecure-by-default plugin architecture forces a centralised marketplace to provide the trust that the platform…
• Pod Disruption Budget — A Pod Disruption Budget (PDB) is a Kubernetes primitive that bounds the number or percentage of pods of a given workload that may be simultaneously terminated during a voluntary…
• Point-checking controls — Point-checking controls is the traditional approach to enforcing privacy or access constraints at a single point of data processing:
• Point-in-Time Recovery (PITR) — Point-in-Time Recovery (PITR) is the database capability of producing a fresh, queryable copy of the database at a chosen past timestamp, typically for:
• Point of presence — A Point of Presence (PoP) is a physical deployment site — typically a data centre (or rack in a colocation facility, or an ISP edge cage) — where a network operator has servers,…
• Poisson sampling for integer outcomes — A trick for converting an ML-predicted expected count into a realised integer count: treat the predicted value λ as the rate parameter of a Poisson distribution and sample count ~…
• Policy-as-data — Policy-as-data stores authorization policies outside application code — in a database, a dedicated policy store, or a config bundle — so they can be changed, audited,…
• Policy bundle — A policy bundle is OPA's distribution unit: a gzipped tarball of .rego policy files + optional data.json reference data, pulled by the OPA agent from a configured remote source…
• Policy lattice (Denning) — A policy lattice is the formal mathematical structure used by information flow control systems to decide whether a given flow of data from source to sink is permitted.…
• Polling interval as freshness budget — When a system coalesces upstream work behind a polling proxy, the proxy's poll interval becomes the upper bound on how quickly any configuration change can reach the downstream…
• POLQA MOS metric — POLQA (Perceptual Objective Listening Quality Analysis, ITU-T P.863) is an objective voice-quality metric that outputs a MOS (Mean Opinion Score) on a 1–5 scale by comparing…
• Polymorphic usage tables (multi-tenant) — Polymorphic usage tables is the application-schema-layer structural decision to use a small fixed set of generic tables — keyed by some shape like (tenantid, entitytype, entityid,…
• Polyrepo shared build logic — The cross-cutting concern an organization with many independent source repositories (vs a single monorepo) has to solve: how do tens of thousands of repos share build conventions,…
• Portable execution environment — A portable execution environment is a uniquely named, fully declared package + binary environment that can be built once and rehydrated anywhere — including at execution time,…
• Position-adaptive diversification — Position-adaptive diversification is a class of feed-diversification algorithms whose decisions condition on items already placed (and, depending on the variant,…
• Positive security model — A positive security model defines what valid traffic looks like and rejects everything that does not conform — the inverse of the traditional negative model (block known-bad…
• Post-quantum authentication — Post-quantum authentication is the migration of digital- signature and credential-verification primitives to quantum- resistant alternatives…
• Post-quantum cryptography — Post-quantum cryptography (PQC) is the class of cryptographic primitives designed to resist cryptanalytic attack by a sufficiently powerful quantum computer.…
• Postgres async I/O — Postgres async I/O is the feature introduced in Postgres 18 (September 2025) that allows the database to issue asynchronous read I/Os to the operating system,…
• Postgres autovacuum — Autovacuum is a Postgres background process that automatically runs VACUUM (and ANALYZE) against tables that have accumulated enough dead tuples to warrant cleanup.…
• Postgres checkpoint — A Postgres checkpoint is a periodic background cleanup event that flushes all modified ("dirty") pages in the buffer pool to disk up to a specific point in the WAL,…
• Postgres failover slot — A Postgres failover slot is the Postgres-17 mechanism (documented as logical replication failover) that serialises logical replication slot state into the WAL so that standbys can…
• Postgres Full Page Write — Full Page Write (FPW) is the Postgres durability primitive that copies the entire 8 KB page into the WAL the first time a page is modified after a checkpoint.…
• Postgres hook — A Postgres hook is a function pointer inside Postgres core that starts out NULL and can be set by an extension to a function that runs before, after,…
• Postgres JSONB vs MySQL JSON — Postgres has two JSON types: JSON (text storage) and JSONB (binary storage, indexable, more compact). MySQL has one: JSON.…
• Postgres logical replication slot — A Postgres logical replication slot is a row in the pgreplicationslots catalog on a Postgres primary that represents a persistent cursor into the WAL from the perspective of one…
• Postgres memory accounting — Postgres memory accounting is the decomposition of a Postgres node's total memory usage into behaviourally-distinct categories that share a display scale but do not share meaning.…
• Postgres MVCC and Heap-Only Tuple (HOT) updates — MultiVersion Concurrency Control (MVCC) in Postgres is implemented by writing a new row version ("tuple") for every UPDATE rather than modifying the existing row in place.…
• Postgres notice warning channel — The Postgres NoticeResponse frame is a wire-protocol message that delivers informational, warning, and debug messages from the server to the driver in-band alongside query results…
• Postgres queue table — A Postgres queue table is a table used as a job queue: workers poll for pending rows, claim one, perform the associated work, and delete (or mark done) the row on commit.…
• Postgres shared_buffers + OS page cache (double buffering) — PostgreSQL does not manage its own full disk-cache hierarchy. Instead, it implements a two-layer caching strategy — its own in-process sharedbuffers cache for recently-accessed…
• Postgres stored-procedure language pluggability — Postgres stored procedures and functions can be written in any registered procedural language — PL/pgSQL (the default), PL/Python, PL/Perl, PL/Tcl, PL/V8 (JavaScript), PL/Ruby,…
• Postgres TRUNCATE CASCADE / RESTART IDENTITY — Postgres's TRUNCATE TABLE is a richer DDL statement than MySQL's. It supports three capabilities MySQL does not: CASCADE, RESTART IDENTITY / CONTINUE IDENTITY,…
• VACUUM FULL (Postgres) — VACUUM FULL is Postgres's compaction mechanism for tables that have accumulated too many dead row versions under MVCC. It purges versions old enough that no running transaction can…
• wal_level=logical — wallevel=logical is the Postgres configuration setting that instructs the primary to include enough information in the WAL to support logical decoding…
• Postgres wire protocol as streaming-SQL surface — The architectural choice of speaking the PostgreSQL wire protocol from a streaming-platform's analytical surface, inheriting the entire Postgres-driver ecosystem at zero…
• postgresql.conf as control plane — Using postgresql.conf (and the reload-without-restart Postgres parameters that live in it) as the control-plane distribution channel to push runtime policy — here,…
• Postmortem as data goldmine — Postmortem as data goldmine is the strategic reframing — canonicalised in Zalando's 2025-09-24 datastore-team postmortem-analysis-pipeline post…
• Power-law URL traffic — Power-law URL traffic is the empirical observation that a small fraction of URLs on a multi-page site receives the overwhelming majority of requests.…
• Power of Two Choices (P2C) — Power of Two Choices (P2C) is a randomised load-balancing primitive that picks two backends uniformly at random for each request and then deterministically routes to the one…
• PQC Migration Levels — PQC Migration Levels is a five-rung maturity ladder proposed by Meta's Security team in 2026-04 for organisations managing post-quantum cryptography migration across many use cases…
• PQC prioritisation framework — The PQC prioritisation framework organises use cases for post-quantum-cryptography migration by attack class, not by asset value. Meta proposes a three-tier taxonomy:
• Practice #4: Everybody gets their own database instance — Practice #4 is the fourth of the seven practices of evolutionary database design (Fowler 2003, Sadalage 2006). Stated cleanly:
• PRAGMA-based point-in-time recovery — PRAGMA-based PITR is a point-in-time-recovery surface that exposes the timestamp knob as a SQL-level PRAGMA on the database connection,…
• Pre-10% fleet detection goal — The pre-10% fleet detection goal is the deploy-safety North Star target that problematic deployments must be detected before reaching more than 10% of the production fleet.…
• Pre-approved degradation procedure — A pre-approved degradation procedure is a mitigation step whose business-impact trade-off has been judged and approved by the business owner of the affected functionality…
• Pre-flight schema-conflict check — A pre-flight schema-conflict check is a conflict- detection pipeline that runs before a deploy request reaches cutover, rejecting the request if it would collide with any other…
• Pre-fork copy-on-write — Pre-fork copy-on-write (COW) is the pattern of a long-running server process *importing / initialising its application state before fork(2),…
• Pre-integration at tier creation — Pre-integration at tier creation names the architectural decision to establish shared dependencies (VPCs, IAM roles, PrivateLink endpoints,…
• Pre-launch Market Validation — Pre-launch market validation is the problem class of assessing product quality for a market (country, locale, audience segment) that is not yet live, i.e.…
• Pre-staged inverse replication — Pre-staged inverse replication is the architectural property that, after an online schema change (or data- motion cut-over) completes,…
• Preallocated memory budget — A performance optimization where a consumer module allocates a fixed-size memory region at startup sized for some upper bound of the data it will process,…
• Precision-aware type mapping — Precision-aware type mapping is the schema-translation discipline where a CDC or data-ingestion connector inspects the source database's precision and scale metadata on numeric…
• Precision-vs-discovery codebook flavor — Precision-vs-discovery codebook flavor is a design axis for Semantic ID codebooks: the same RQ-VAE quantizer + contrastive-loss machinery can be trained against different upstream…
• Precomputed predictions API — A precomputed predictions API is the ML-serving pattern where a scheduled batch job fills a low-latency key-value store with all (or most) predictions ahead of time,…
• Predicate locking (Postgres) — Predicate locks are Postgres's non-blocking locking primitive for tracking the row-sets (not the individual rows) a transaction has touched,…
• Predictive auto-scaling — Predictive auto-scaling is capacity control that scales a workload's resources before observed load arrives, using a forecast of future demand,…
• Preference schema decoupling — Preference schema decoupling is the structural discipline of representing a user preference as a tuple of orthogonal axes rather than a single conflated enum,…
• Prefetch-window metadata co-attending — Prefetch-window metadata co-attending is the client-infrastructure primitive of piggybacking a new piece of per-item metadata onto the existing prefetch path for that item,…
• Prefill/decode (PD) disaggregation — Prefill/decode disaggregation (or PD disaggregation) is an LLM-serving architectural pattern that separates the two stages of LLM inference…
• Prefix-aware routing (LLM inference) — Prefix-aware routing is the inference-request routing strategy that sends requests sharing a common prompt prefix to the same model-serving replica,…
• Preheating — Preheating is a cache-population discipline in which a high-fanout source surface (issue list, dashboard, project view) proactively walks references to likely-next destinations…
• Prepared statement memory accumulation — Prepared statement memory accumulation is the per-session growth of cached query plans and prepared statements that is not released until the session ends or the statement…
• Presentation policy — A presentation policy in Zalando's Interface Framework is a named, valued directive inside an Experience that tells a domain backend how to shape its GraphQL response…
• Price-performance ratio — The price-performance ratio (equivalently: cost per unit of performance, $/QPS, $/TPS, cost-per-IOPS) is the consumer-facing metric that answers "how much does it cost to achieve…
• Primary-Replica Topology Alignment — Primary-replica topology alignment is the structural principle that the replication topology of your storage layer should mirror the write-ownership topology of the application…
• Primary-standby failover — Primary-standby failover is the operational move of promoting a previously-passive standby cluster (or node, or region) to the primary role when the current primary becomes…
• Primary vs replica as replication source — When a backup process, migration tool, new replica, or CDC consumer needs to replicate from a database cluster, it has a choice of source…
• Priority Flow Control (PFC) — Priority Flow Control (PFC), defined in IEEE 802.1Qbb, is a link-level flow-control mechanism that lets a switch tell its upstream neighbour to pause traffic per-traffic-class (up…
• Proactive cache of batch predictions — Proactive cache of batch predictions is a serving strategy where a batch job precomputes and stores predictions for all relevant entities before any user requests them,…
• Probabilistic data structure — A probabilistic data structure trades exact correctness for sub-linear space or time by returning approximate answers with known error bounds.…
• Probabilistic demand forecast — A probabilistic demand forecast is a forecast whose output is not a single point estimate but a distribution over possible future demand values…
• Probabilistic record linkage — Probabilistic record linkage is the formal-statistical formulation of entity resolution: given two records across (or within) datasets,…
• Problem vs Error distinction — In Zalando's 2021 GraphQL error-modeling discipline, Problem and Error are distinct, non-overlapping vocabulary:
• Process (OS) — A process is an instance of a program being executed by an operating system. It is the OS's fundamental unit of isolation for running code: each process has its own address space…
• Producer backpressure batch growth — Producer backpressure batch growth is the counterintuitive behaviour of Kafka-API producers under broker saturation: when the broker is heavily loaded,…
• Producer-consumer loop — A pipeline architecture where later stages (Gapfill, Feedback, Trace) continuously produce new tasks while earlier stages (Validate, Dedup, Report) consume them,…
• Product catalog syndication — Product catalog syndication is the commerce problem of getting a seller's product catalog into every buy-side surface that wants to list it — in the shape each surface demands.…
• Product-market fit — Product-market fit (PMF) — the well-known startup term for the state in which a product has found a customer segment that actively wants it and will pull it from the company,…
• Production data diversity — Production data diversity is the observation that the distribution of inputs an ML model sees in production is systematically broader and messier than the distribution it sees…
• Production readiness review — A Production Readiness Review (PRR) is an SRE-book-named structured audit of a service's reliability posture, performed before the service takes on a new reliability risk — e.g.…
• Production traces as evaluation substrate — Production traces as evaluation substrate is the architectural property of using durable, queryable production traces as the source of truth for evaluation datasets…
• Production-version cloning for load test — Production-version cloning for load test is the practice of programmatically determining the exact versions currently deployed in production for every application in scope…
• Profile-guided optimization — Profile-guided optimization (PGO) is a compiler-level optimisation technique where the compiler consumes profile data collected from a running instrumented binary (or…
• Program correctness — Program correctness is the property that a program produces the intended result for every valid input, in every reachable state, under every allowed interleaving.…
• Progressive capability disclosure — Progressive capability disclosure is the design principle that an agent (or any consumer) should discover a system's capabilities incrementally, at the moment they are needed,…
• Progressive delivery per database — Progressive delivery per database is the fleet-rollout discipline for a database-as-a-service vendor where configuration and binary changes ship one customer database at a time,…
• Progressive hydration — Progressive hydration (sometimes partial hydration when only some subtrees hydrate at all, or selective hydration when priority is dynamic) is the property that a page's…
• Progressive screen-level RN migration — Progressive screen-level RN migration is the adoption mode for a brownfield React Native rollout: migrate one screen at a time, each on its own risk budget,…
• Progressive tool disclosure — Progressive tool disclosure is the practice of exposing an LLM agent's available tools on demand rather than loading every tool schema into context upfront.…
• Project rules / steering files — Project rules (also "steering files", "agent rules") = Markdown or config files checked into the repository that describe architectural constraints and coding conventions the AI…
• Projection layer for latency — A projection layer for latency is a learned layer inserted between an expensive upstream encoder (typically a wide Transformer) and expensive downstream layers (feature crossing,…
• Promise allocation overhead — Promise allocation overhead is the per-call CPU + memory cost of creating, resolving, and garbage-collecting Promise objects in a JavaScript hot path.…
• Promise pipelining — Promise pipelining is an RPC technique in which the caller uses the result of an in-flight call as an argument to — or receiver of — a subsequent call,…
• Prompt-boundary sanitization — Prompt-boundary sanitization is the practice of stripping any occurrence of the structural delimiters of an LLM prompt from user-controlled content before concatenating…
• Prompt-cache consistency — Prompt-cache consistency is the design constraint of keeping the prefix of a prompt stable across requests — even when parts of it must be dynamic…
• Prompt chaining — Prompt chaining is the LLM prompting technique in which an application *programmatically threads the assistant's response at step N into the user prompt at step N+1,…
• Prompt injection — Prompt injection is an adversarial attack against an LLM where attacker-controlled text, embedded in input the LLM is expected to process,…
• Prompt interface+mapping+examples composition — Interface+Mapping+Examples is the three-layer prompt composition Zalando's Partner Tech department converged on for LLM-powered code migration between two UI component libraries.…
• Prompt is not control — Prompt is not control is the architectural principle that fine-grained control over agent behaviour cannot be achieved via prompt instructions alone.…
• Prompt optimisation feedback loop — A prompt optimisation feedback loop is a production pattern in which LLM prompts are continuously edited based on real production outputs,…
• Prompt prefix caching — Prompt prefix caching is a cost and latency optimization where LLM providers reuse the KV-cache from a previous inference call if the current call's prompt shares an identical byte…
• PromQL-to-SQL translation — A translation layer that accepts PromQL queries from existing dashboards / alerting rules / Grafana panels and executes them as SQL against a columnar-object-storage lakehouse…
• Proof of exploitability — A proof of exploitability is runnable code that demonstrates a suspected vulnerability is actually exploitable — closing the gap between "I found a flaw in the code" and "I have…
• Property graph — The property graph is a graph data model where:
• Proposal number — A proposal number is a totally-ordered, timestamp-like identifier attached to every leadership-change attempt in a lock-free consensus algorithm.…
• Proprietary database operator — A proprietary database operator is a managed-database vendor's closed-source control plane — the provisioning, failover, backup, replication,…
• Proto3 explicit optional — Protobuf 3's three-way history with field presence: proto2 had required and optional labels; proto3 (initial) dropped both…
• Provider failure taxonomy — A provider failure taxonomy is a classification of the distinct failure modes an external API can return, enabling the caller to choose a class-specific retry policy instead…
• Provider service catalog API — A provider service catalog API is a machine-readable, JSON-over-REST feed that a cloud provider publishes listing every service an agent can provision through the provider.…
• Provisioned Throughput vs On-Demand (LLM serving) — Provisioned Throughput (PT) and On-Demand (OD) are the two canonical capacity primitives that a managed LLM serving substrate (Amazon Bedrock, GCP Vertex AI, Azure AI Foundry,…
• Proxy-node correlation signal — A proxy-node correlation signal fires when the same session identifier — typically a browser-telemetry fingerprint — appears from multiple proxy-node IP addresses within a window…
• Proxy tier latency tax — The proxy tier latency tax is the additional per-query network hop incurred when the application talks to a proxy (e.g. Vitess VTGate, PgBouncer,…
• Proxy-transparent failover — Proxy-transparent failover is the user-facing property that when a database node is replaced (due to crash, promotion, upgrade, or scale event),…
• Pseudo-context augmentation — Pseudo-context augmentation is a training-time data-augmentation technique used when a model needs to learn to consume a request-time-only feature that is, by construction,…
• Psychoacoustic compression — Psychoacoustic compression exploits properties of the human auditory system — masking, perceptual thresholds, and the specific statistical structure of human voice…
• Public ID column — A publicid column is a VARCHAR(N) column holding an externally-visible opaque identifier, separate from the table's internal numeric primary key. The column is:
• Public-to-private routing — Routing traffic from the public Internet to applications running on private networks (RFC 1918/6598/4193 address space) through an intermediary proxy that terminates the public…
• Publish latency as congestion signal — Publish latency as congestion signal is the use of a producer-observed message-broker publish time — typically the P50 (median) of the round-trip from producer publish() to broker…
• Publish-time immutability — Publish-time immutability is the semantics in which a composite software artifact — a release, a package version, a container tag…
• Pull vs push streams — A streaming API is fundamentally either pull-based (the consumer drives — data advances when the consumer asks) or push-based (the producer drives…
• Purpose limitation — Purpose limitation is a core data-protection principle: personal data is only processed for explicitly stated, legitimate purposes and not further processed in ways incompatible…
• Push aggregation under join — Pushing aggregation under a join is a query-planner rewrite that moves a GROUP BY + aggregate operator from above a join (where it sees the cartesian / nested-loop product of both…
• Push-based invalidation — Push-based invalidation is a reactive-update strategy: maintain an explicit dependency graph from sources to dependents; when a source value changes, mark all dependents dirty;…
• Push vs pull monitoring — A monitoring architecture is fundamentally either pull-based (the monitoring server periodically scrapes metrics from endpoints) or push-based (the monitored system pushes metrics…
• Pydantic structured LLM output — Pydantic structured LLM output is the Python-ecosystem convention of declaring a Pydantic BaseModel for every LLM output shape and treating the model's JSON response as either…
• PySpark UDF for Complex Business Logic — A PySpark User-Defined Function (UDF) wraps a Python function so it can be applied row-wise (or grouped-row-wise) inside a Spark DataFrame operation. Two flavours:
• Q-Day — Q-Day is the day a cryptographically-relevant quantum computer (CRQC) can break the asymmetric cryptography actively protecting deployed systems — RSA, classical Diffie-Hellman,…
• QLC read-vs-write asymmetry — QLC read-vs-write asymmetry is the property — intrinsic to Quad-Level-Cell NAND physics — that reads complete substantially faster than writes on the same media.…
• qlog QUIC instrumentation — qlog is the standardised JSON event-log format for QUIC and HTTP/3, maintained as an IETF QUIC WG draft and adopted by the major open-source QUIC implementations (including…
• Quality / bitrate / complexity tradeoff — Audio codec design is governed by a three-axis tradeoff: increasing quality generally requires either more bitrate (more bytes on the wire) or more complexity (more CPU per second…
• Quality penalty signal — A quality penalty signal is an upstream classifier output that flags individual content items as belonging to a sensitive or elevated-risk class…
• Quantization — Quantization rescales tensor elements from a high-precision floating-point range into a smaller number of discrete levels represented with fewer bits.…
• Quantum Key Distribution (QKD) — Quantum Key Distribution (QKD) is a family of key-agreement protocols that uses quantum physical phenomena — typically the transmission of polarised photons over a dedicated…
• Quasi-experimental methods — Quasi-experimental methods are causal-inference techniques for estimating treatment effects when randomization at the level of interest is infeasible.…
• Query buffering cutover — Query buffering cutover is the technique of briefly holding ("buffering") incoming client queries at a proxy layer while the backend behind the proxy is swapped from one database…
• Query consolidation — Query consolidation is the database-proxy primitive where multiple identical SELECT queries arriving nearly simultaneously are deduplicated at the proxy tier: exactly one upstream…
• Query digest — A query digest is a normalised form of a SQL statement with literal values stripped, used as the unit of workload aggregation in observability tooling.…
• Query-frequency power-law caching — Query-frequency power-law caching is the concrete infrastructure primitive that exploits a power-law distribution over queries to dramatically shrink the set of inputs you have…
• Query history as knowledge base — Query history as knowledge base is the framing shift of treating accumulated SQL query logs — normally throwaway audit data — as a durable,…
• Query lifecycle manager — A query lifecycle manager is the component of a distributed SQL engine "responsible for the lifecycle of currently-running queries" — scheduling them onto executors,…
• Query parameter pattern — A query parameter pattern is the sorted set of parameter names present in a URL's query string, ignoring values. It is the grouping key that makes parameter classification…
• Query pattern fingerprint — A query pattern fingerprint is a stable hash of a normalised SQL statement — the grouping key that tells an observability system "these two executions are the same pattern." Two…
• Query planner — A query planner (also query optimizer) is the component of a database system that takes a textual query — SQL or another query language…
• Query preprocessing (tokenization, normalization, rewriting) — Query preprocessing is the pre-retrieval pipeline stage that transforms a raw user query into a clean form suitable for downstream retrieval. Three canonical sub-steps:
• Query priority classification — Query priority classification is the upstream step that makes load-shedding feasible: every database query is tagged — at the application layer…
• (query, product) Evaluation Cache — The (query, product) evaluation cache is the deduplication pattern scoped to an offline evaluation pipeline: cache every (queryid,…
• Query shape — A query shape is the un-parameterized form of a query, assigned a stable unique ID. A live query is then identified by (shapeid, argumentvalues).…
• Query simplifier — A query simplifier is a tool that takes a SQL query that exhibits a specific error (wrong result, crash, planner panic, resource exhaustion) and automatically produces the smallest…
• Query size derived from the logical plan — Query size derived from the logical plan is the primitive of estimating the resource cost of a query from its optimiser-internal representation, before execution,…
• Query stage profiling — Query stage profiling is a MySQL diagnostic technique that decomposes a single query execution's wall-clock time into per-stage milliseconds: how much was spent optimising,…
• Query-statistics telemetry — Query-statistics telemetry is the practice of tracking per-query execution statistics — count, rows returned, duration — at the query-path proxy layer,…
• Query-structure-aware caching — Query-structure-aware caching is caching where the cache parses the incoming query, understands its logical structure, and decomposes the response along a structural axis…
• Query tag filter — A query tag filter is the search-syntax primitive in a database-observability UI (or query-log search tool) that scopes results to queries or errors carrying a named tag,…
• Query telemetry as deploy-safety signal — Query telemetry as deploy-safety signal is the architectural pattern where the database platform uses its own runtime workload observations…
• Query understanding — Query understanding (QU) is the upstream stage of a search pipeline that turns a raw user query string into structured intent signals downstream retrieval + ranking can consume.…
• Query vs document embedding (two distinct serving problems) — In retrieval / search / recommendation systems that use vector embeddings, the embedding- inference workload splits into two sub-populations with fundamentally different serving…
• query_wait_timeout — querywaittimeout is the PgBouncer configuration that governs admission-control behaviour when the pool is exhausted: how long a client waits for a connection before being…
• Queue-admission conflict warning — A queue-admission conflict warning is a schema-change workflow feature that validates a newly-submitted deploy request against all currently-queued changes at submission time,…
• Queue depth as latency-hiding mechanism — Queue depth as latency-hiding mechanism is the architectural specialisation of Little's Law in which a queue is added to a pipeline stage not to absorb temporary arrival-rate…
• Queue length vs wait time — For any queue, there are two natural observables:
• Queueing theory (as applied to storage/IO stacks) — Queueing theory is the math of how waiting lines form and drain when arrivals are asynchronous. Applied to systems: between the CPU and durable storage there is always a chain…
• QUIC connection migration — QUIC connection migration is the QUIC-specific primitive that lets a client change network path — e.g. hand over from Wi-Fi to cellular,…
• QUIC-TLS fused handshake — The QUIC-TLS fused handshake combines transport-parameter negotiation and TLS 1.3 cryptographic handshake into a single round trip.…
• QUIC transport — QUIC is a user-space, UDP-based transport protocol standardised by the IETF as RFC 9000 (May 2021). Unlike TCP — a kernel-space,…
• Quickening (runtime bytecode rewrite) — Quickening is an interpreter optimisation in which the VM rewrites its own bytecode at runtime to replace generic opcodes with type-specialized variants once operand types have…
• Quorum read — A quorum read is a read that contacts a majority (or durability-defining quorum) of followers to determine the most recent durable value.…
• Quota auto-rightsizing — Quota auto-rightsizing is the periodic, automated adjustment of a quota's numeric value based on that subject's historical usage pattern…
• Quota lifecycle management — Quota lifecycle management is the discipline of treating a quota as a first-class object with a full CRUD + operational lifetime — schema definition, validation, authorization,…
• RA-TLS (Remote Attestation + TLS) — RA-TLS is the composition in which a TLS handshake carries (or piggybacks) a fresh remote attestation of the server TEE's binary digest + instance identity,…
• Rack-level power density — Rack-level power density is the amount of electrical power a single datacenter rack can deliver and dissipate, measured in kW per rack.…
• Radiation effects on computing — Radiation effects on computing is the failure-mode class introduced when silicon operates in an environment with meaningfully higher ionising-radiation flux than terrestrial sea…
• Radio-access-network bottleneck — In the end-to-end Web protocol stack, the radio-access network (RAN) — the physical link between a mobile device and the carrier's cell tower…
• Raft-replicated WAL as LSM WAL — Instead of maintaining a traditional file-based write-ahead log for an LSM-tree database, the Raft consensus log is used directly as the WAL.…
• RAG as a judge — RAG as a judge is the pattern of letting an LLM judge fetch its own context — via retrieval — before scoring a candidate response,…
• Rails query log tags — Rails query log tags are Rails 7+'s native ActiveRecord feature for automatically appending a trailing SQL comment to every ActiveRecord-issued query,…
• Random instance failure injection — Random instance failure injection is the primitive operation of chaos engineering: pick a production instance at random and kill it,…
• Range partitioning — Range partitioning is a MySQL partitioning strategy that assigns rows to partitions based on whether a column value falls inside a pre-declared numerical range.…
• Range sharding — Range sharding routes each row to a shard by whether its shard-key value falls inside a pre-defined range assigned to that shard.…
• Ranked rack preferences — Ranked rack preferences is the Redpanda 26.1 extension to leader pinning that turns the leader- placement hint from a single preferred location into an ordered list of preferred…
• Ranking via election — Ranking via election is a tournament-style prompt-structure pattern for applying an LLM ranker to a candidate set larger than any single prompt can hold.…
• Rate-limit trilemma — The rate-limit trilemma is Cloudflare's framing for a fundamental tension in how the Internet governs access: a web-access-governance system can be any two of decentralized,…
• Rate-limited cache — A rate-limited cache is a lookup cache in front of an expensive or rate-sensitive backend, with an explicit cap on lookup rate per key (not just entry lifetime). The shape is:
• RDF (Resource Description Framework) — RDF — Resource Description Framework — is the W3C standard graph data model for representing information as triples: (subject, predicate, object).…
• RDMA KV transfer — RDMA KV transfer is the serving-infrastructure primitive of moving LLM KV cache blocks between GPUs (intra-node) or nodes (inter-node) using Remote Direct Memory Access protocols…
• RDMA Queue Pair — In RDMA (both InfiniBand and RoCE), a Queue Pair (QP) is the fundamental transport-endpoint abstraction: a paired Send Queue + Receive Queue on each side of a connection,…
• Reachability-based subscription — Reachability-based subscription is a server-side real-time collaboration pattern in which a session's subscription to document edits is defined as the transitive closure…
• React custom reconciler — A React custom reconciler reuses React's component model, element tree, and reconciliation (diffing + commit) algorithm while replacing the host tree…
• React Hydration — Hydration is the process by which a frontend React runtime takes over server-rendered HTML: it walks the existing DOM, verifies the markup matches what it would render locally,…
• React Native as a package — React Native as a package is an architectural pattern for integrating React Native into an existing large native iOS/Android codebase without requiring a full-app rewrite…
• React re-render — React re-render is the cost of React re-invoking a component's render function + diffing the returned virtual-DOM tree + reconciling to the real DOM when state or props change.…
• Reactive auto-scaling — Reactive auto-scaling is capacity control that observes over- or under-utilisation and then changes capacity — the standard industry default from 2010s-era AWS Auto Scaling Groups…
• Read amplification — Read amplification is the ratio between the number of client-issued read operations and the number of substrate-level read operations they fan out into.…
• Read committed isolation — READ COMMITTED is the ANSI-SQL isolation level at which each SELECT in a transaction reads the latest committed view of the database.…
• Read-invalidation rendezvous — Read-invalidation rendezvous is the concurrency problem of coordinating in-flight reads against in-flight invalidations in an invalidation-based cache so that no invalidation…
• ReadPartialCommits (proposed isolation level) — ReadPartialCommits is a proposed isolation level — not shipped in any production database as of the 2026-04-21 ingest — introduced by Sugu Sougoumarane in his 2020 canonical…
• Read-recursive lock — A read-recursive lock is a RwLock acquisition API that lets a thread acquire a read lock even if it already holds one, bypassing the writer-preference starvation avoidance logic…
• Read-side detection of storage pathology — Read-side detection of storage pathology is the design choice to detect storage-engine pathologies (wide partitions, hot keys, oversized blobs,…
• Read-through NVMe cache (sparse) — A disposable local-NVMe cache sitting in front of an immutable object-store-backed disk. The volume is sparse (backing store grows only as chunks are fetched),…
• Read-time preference translation — Read-time preference translation is the discipline of migrating a preference schema's semantics without touching the stored bytes,…
• Read uncommitted isolation — READ UNCOMMITTED is the lowest ANSI-SQL isolation level. Transactions see the latest value of every row — including values written by other transactions that have not yet…
• Read-write splitting — Read-write splitting is the application-level discipline of directing reads to one or more replicas and writes to the primary in a replicated database cluster.…
• Read-your-writes consistency — Read-your-writes consistency (RYW) is the session-consistency guarantee that, within a single client session, any read issued after a successful write observes a state…
• Real-time context feature — A real-time context feature is a feature whose value is determined at request time by the immediate session state — the page a user is currently viewing,…
• Real Time Factor (RTF) — Real Time Factor (RTF) is the ratio of processing wall-clock time to output media duration:
• Real User Measurement — Real User Measurement (RUM) is performance data collected from real end-user browsers / devices in production traffic, as opposed to synthetic measurements from dedicated test…
• Rebuffering rate — Rebuffering rate is the frequency (interruptions per streaming session, per viewing hour, or per 100 view-hours) at which a video-streaming client runs its playback buffer empty…
• Rebuild-Restore-Rotate framework — Rebuild-Restore-Rotate is the cyber-event recovery decomposition framework that sorts every recovery target into one of three categories with a one-line per-category heuristic:
• Receiver-driven traffic admission — Receiver-driven traffic admission is a congestion-control discipline in which the receiver (not the sender) decides when traffic may enter the network.…
• Reciprocal active-passive clusters — Reciprocal active-passive clusters is a multi-region streaming architecture where two clusters each act simultaneously as a source cluster for their own data and as a shadow…
• Reciprocal Rank Fusion (RRF) — Reciprocal Rank Fusion (RRF) is a score-fusion technique for combining result lists from multiple independent retrieval methods into a single ranked list,…
• Reclaimable vs non-reclaimable memory — Reclaimable memory is memory the kernel can transparently drop when another process requests pages — page cache, buffer cache, clean memory-mapped pages backed by files on disk.…
• Reconstruction-vs-semantic loss tradeoff — In representation-learning architectures with multiple loss-term objectives, the reconstruction-vs-semantic loss tradeoff is the tension between fidelity (the encoder/decoder…
• Recursive CTE — A recursive CTE (recursive common table expression) is the SQL WITH RECURSIVE name AS (basequery UNION [ALL] recursivereference) construct that lets a query reference its own…
• Recursive entity tree resolution — Recursive entity tree resolution is the tree-walk algorithm implemented by entity-based page composition runtimes: take a root Entity,…
• Redshift Connector latency — Redshift Connector latency is the name for the multi-hour delay between "the data pipeline has produced output" and "that output is queryable in Redshift" when the data path runs…
• Redundant condition (query hint) — A redundant condition is a WHERE clause predicate that is logically redundant — it does not change the result set — but is added deliberately so that the query planner can pick…
• Redundant error signalling — Redundant error signalling is the UI-design discipline of encoding error state across multiple perceptual channels rather than relying on any single one — classically,…
• Reference table materialization — Reference table materialization is the technique of copying a small, read-mostly lookup table from an unsharded keyspace into every shard of a sharded keyspace,…
• Refinement-round budget — Refinement-round budget is the bounded-iteration discipline of a judge-gated agent loop: every loop has a hard ceiling on the number of plan → implement → verify → refine cycles it…
• reflect.MethodByName linker pessimism — reflect.MethodByName(name) with a non-constant name is the single Go language feature that most aggressively defeats the linker's method dead-code elimination.
• Regex + enum validation — Regex + enum validation is the two-layer validation discipline in which:
• Region-split experiment — A region-split experiment applies an intervention to a whole set of geographic markets (the treated regions) and compares their outcome trajectory to a set of untreated markets…
• Regional read replica — A regional read replica is a full, read-only copy of a primary database's dataset, placed in a geographic region distinct from the primary region,…
• Regionalization blast-radius reduction — Regionalization is the structural response to a single global state-distribution cluster whose blast radius has proven unacceptable: split the state into a per-region cluster (with…
• Register-based client capability matching — Register-based client capability matching is the server-side mechanism that decides, per feature, whether the requesting client can render that feature,…
• Registry as governance precondition — Registry as governance precondition is the architectural argument that an inventory primitive is upstream of any policy-layer enforcement: a registry doesn't do governance,…
• Registry-less schema evolution — Registry-less schema evolution is the property of a streaming sink connector to evolve the downstream table's schema directly from the shape of raw incoming records…
• Reified edge graph — A reified edge graph is a data-modeling pattern where relationships between entities are stored as first-class records (facts) rather than as graph-DB-native edges or as values…
• Reindex-based cluster upgrade — A reindex-based cluster upgrade is a stateful-datastore major- version upgrade strategy where, instead of upgrading the live cluster node-by-node (rolling upgrade),…
• Relative Score Fusion (RSF) — Relative Score Fusion (RSF) is a score-fusion technique for combining result lists from multiple independent retrieval methods into a single ranked list,…
• Release attestation — A release attestation is a signed cryptographic statement that a specific release — identified by tag, asset digests, and optionally build-pipeline metadata…
• Release-channel rollout — Release-channel rollout is a deployment discipline in which a change must graduate through an ordered sequence of channels before reaching production.…
• Relevance labeling — Relevance labeling is the activity of assigning a graded score to a (query, document) pair that encodes how well the document satisfies the query.…
• Reliability-velocity duality — Reliability-velocity duality is the engineering philosophy that reliability and velocity of innovation are not opposing forces but "two facets of the same coin"…
• Remote attestation — Remote attestation is the cryptographic mechanism by which a client can verify — from outside — that a specific, known-good software image is running inside a genuine TEE…
• Remote Build Execution (RBE) — Remote Build Execution (RBE) distributes build/test actions across a cluster of worker machines managed by a central build service. The client (e.g.…
• Remote development environment — A remote development environment is an architectural setup where the developer's editor UI runs on one machine (typically a laptop) but code execution, language services,…
• Remote read replica topic — A remote read replica topic is a read-only topic on a separate cluster that mirrors a topic on an origin cluster by reading the origin's tiered-storage / archival-storage segments…
• Remove before add optimization — Remove before add optimization is the architectural principle of auditing existing optimisation choices for measured benefit before adding new ones.…
• Render-as-you-fetch — Render-as-you-fetch is the React-native data-loading pattern that concurrent rendering is designed for. A component renders, its data-fetching hook initiates the fetch,…
• Rendering delay distribution (Google, 2024) — The rendering delay distribution is the canonical empirical percentile distribution of time between Googlebot's initial crawl of a page and the Web Rendering Service's completion…
• Rendering queue (Google) — The rendering queue is the FIFO-ish staging area between Googlebot's crawl stage (HTTP fetch of the URL) and its render stage (execution by the Web Rendering Service).…
• Rendering-strategy crawl-efficiency tradeoff — The rendering-strategy crawl-efficiency tradeoff is the observation that different rendering strategies (SSG / ISR / SSR / CSR) produce materially different Google-facing…
• Replay training — Replay training (also: "rehearsal", "experience replay" in a related RL sense) is the technique of including examples from a model's previous training distribution in the new…
• Replica-cost tradeoff — Replica-cost tradeoff is the observation that replica count does not scale linearly with availability SLA — doubling replicas does not double durability or availability,…
• Replica creation from backup — Replica creation from backup is the production technique of seeding a new replica by restoring a recent backup onto a fresh empty server and then catching up via replication…
• Replication heartbeat — A replication heartbeat is a timestamp row periodically written to a dedicated table on the replication primary; the same row arrives on replicas via normal replication,…
• Replication lag in message count — Replication lag in message count is the measurement dimension where cross-cluster replication lag is reported as the number of source-cluster messages not yet replicated…
• Replication lag — Replication lag is the time between a write being applied on the primary server of a replicated database cluster and that same write becoming visible on a replica (or secondary /…
• Repo-per-agent-session — Repo-per-agent-session is the pattern-concept of giving every agent session its own Git repository that persists the session's filesystem state + prompt history + any other…
• Request cancellation — Request cancellation is the elector-initiated action of deleting a tentative request across followers during a leadership change,…
• Request collapsing — Request collapsing is the CDN/cache behaviour in which N concurrent requests for the same uncached (and cacheable) resource are deduplicated into a single upstream invocation: one…
• request_concurrency as autoscaling signal — requestconcurrency is the count of in-flight requests being served by a pod at any given instant. As an autoscaling signal it sits in the family of reactive metrics…
• Request-level deduplication — Request-level deduplication is the family of techniques that store, transform, train on, and serve request-level data once per request,…
• Request-level embedding broadcast — Request-level embedding broadcast is a serving-side optimisation for ranking / CTR models where a heavy per-entity embedding (typically the user embedding) is fetched once per…
• Request-oriented computation sharing — Request-oriented computation sharing is the architectural shift in recsys serving from per-(user, ad-candidate) independent inference to per-request shared computation.…
• Request-oriented sequence scaling — Request-oriented sequence scaling is the storage-and-compute architecture that unlocks the use of long-form user behaviour sequences in a ranking model by (1) processing…
• Request propagation — Request propagation is the concern of surfacing requests that were durable under a prior leader onto the new leader's follower set during a leadership change,…
• Request queue depth metric — Request queue depth — the number of in-flight requests waiting inside a service — is a saturation leading indicator that frequently localises an upstream bottleneck when rate…
• Request timeout — A request timeout is the maximum time a client waits for a response from a server after the TCP connection has already been established.…
• Request versioning — Request versioning is the assignment of a totally-ordered, time-based version to every in-flight request in a consensus system,…
• Reserved admin connection budget — The reserved admin connection budget is the operational discipline of never letting PgBouncer consume all of Postgres's maxconnections…
• Reserved connection — A reserved connection is a database connection that has been taken out of a shared connection pool and dedicated to a single caller's session for the session's lifetime.…
• Residualized regression — Residualised regression is the estimation stance of regressing deviations from a learned baseline on deviations of the explanatory variables from their baseline…
• Resilient inference stream — A resilient inference stream is an LLM-inference response stream whose lifetime is decoupled from the lifetime of the caller that issued it.…
• Resource annotations (MCP) — Resource annotations are optional metadata attached to an MCP resource registration that let clients filter, sort, and display resources sensibly at discovery time.…
• Resource enforcement bypass via SSH — When jobs are executed via direct SSH onto a cluster's master node, they bypass the cluster resource manager's enforcement of per-job memory/CPU limits.…
• Resource link vs embedded resource — Resource link vs embedded resource is the token-budget / bandwidth trade-off that emerges when an MCP tool returns resource-typed content.…
• Resource Record Set — A Resource Record Set (RRset) is a DNS protocol primitive defined in RFC 1034 §3.6: a collection of records that share the same owner name, type, and class.…
• Resource stranding — Resource stranding is the failure mode where a server's remaining resources are too small or unbalanced to host any candidate workload, effectively wasting capacity.…
• Resource subscribe capability (MCP) — The resource subscribe capability is the MCP-protocol edge that lets an MCP server push notifications/resources/updated events to connected clients whenever a registered resource's…
• Response-body sampling — Response-body sampling is the technique of inspecting a uniformly-random subset of outbound API responses — parsing the actual response body,…
• Response comparison — headers, body, status — Response comparison along three axes is the canonical diff framework used in API-level parallel runs: two HTTP responses are compared along HTTP status, the headers set,…
• REST-based job submission — REST-based job submission is the architectural model in which a long-running job's lifecycle is managed as a resource addressable by a job ID over HTTP,…
• Retailer-partitioned index — A retailer-partitioned index is the post-decode mapping layer in a multi-tenant generative retrieval system: an index keyed by (retailerid,…
• Retirement + lifespan + buffer formula — The retirement + lifespan + buffer formula is the simple arithmetic rule that decides when a retired signing key's public half can be safely dropped from the JWKS endpoint:…
• Retrieval-Augmented Generation (RAG) — Retrieval-Augmented Generation (RAG) is the inference-time architectural pattern where an LLM's context is augmented with documents retrieved from an external knowledge base…
• Retrieval → ranking funnel — The retrieval → ranking funnel is the canonical two-stage architecture for recommendation, search, and recommendation-like systems at scale:
• Reuse existing infrastructure over purpose-built service — Reuse existing infrastructure over purpose-built service is the heuristic that says: when the platform you already operate can do the job adequately,…
• Revenue Recognition Automation — Revenue Recognition is the accounting principle that determines when and how much revenue a company books from a sale — distinct from when the cash changes hands.…
• Reverse leaky bucket — A reverse leaky bucket is a rate-limiter topology in which each admitted unit of work accumulates debt in the bucket (rather than consuming credits),…
• Reverse-order revert — Reverse-order revert is the property that undoing a multi-migration schema deployment by applying each migration's revert in the inverse of the original deploy order mathematically…
• Reverse replication workflow — A reverse replication workflow is a second replication stream, flowing from the new system back to the old system, created at the moment of cutover and kept running after…
• Reverse shadow phase — The reverse shadow phase is the phase of a Shadow → Reverse Shadow → Cleanup migration in which the roles of the two jobs swap: the previously-shadow job is promoted to write…
• Reversible transform sequence — A reversible transform sequence is the pipeline of operations a format-aware compressor applies to input data before entropy coding.…
• Review-highlight phrase expansion — Review-highlight phrase expansion is the LLM-driven task of generating a creatively-expanded list of phrases that are semantically adjacent to a user's search query,…
• Revocation feed subscription — A revocation feed subscription is a cache-invalidation strategy for token systems: instead of replicating a global blacklist to every verifier,…
• Revoke-and-establish split — The revoke-and-establish split is the pedagogical decomposition, from Sugu Sougoumarane's Consensus algorithms at scale series,…
• RFC as codified engineering rule — A discipline where an engineering organisation distils RFCs (Requests For Comments) — long-form design documents written by domain experts to establish a best practice…
• RFC normative language — RFC normative language is the set of capitalised keywords — MUST, MUST NOT, SHOULD, SHOULD NOT, MAY, etc. — that Internet standards use to express requirement strength.…
• RIB / FIB routing split — In networking, the RIB (Routing Information Base) is the router's full, authoritative routing table — built from every route the router has learned from every protocol.…
• Ride cancellation rate — Ride cancellation rate is the share of dispatched rides that are cancelled after driver match but before trip completion.…
• Rider-driver communication black hole — A rider-driver communication black hole is the anti-pattern in ride-sharing / on-demand-logistics apps where last-mile coordination is offloaded to ad-hoc chat/call exchanges…
• Risk-tier assessment — Risk-tier assessment is the discipline of classifying each code change into a small number of risk buckets before deciding how much review firepower to apply to it.…
• RLHF as offline batch — Reinforcement Learning from Human Feedback (RLHF) is the named fine-tuning pipeline that — as of 2026 — sits on the batch side of the frontier-model batch-training boundary.…
• Robot Experience (RX) — Robot Experience (RX) is the product-design axis that sits alongside UX (end-user experience) and DX (developer experience) and asks: when an LLM-driven coding agent ("robot") uses…
• robots.txt compliance — robots.txt compliance is the practice — by a crawler operator — of fetching /robots.txt from every origin before crawling, and honoring the Disallow, Allow,…
• robots.txt — robots.txt is a text file at the root of a site (https://example.com/robots.txt) that declares crawl rules for automated clients — which user-agents may access which paths.…
• Role prompting — Role prompting is the LLM prompt-engineering technique of assigning the model a specific persona, domain role, or expertise frame — typically via the system prompt…
• Rolling upgrade — A rolling upgrade is an upgrade idiom where the units of a fleet (containers, tablets, hosts, pods) are upgraded one at a time rather than all at once — each unit is drained,…
• Rolling-window query — A rolling-window query is a time-series query whose window advances continuously with clock time — typically [now - Δ, now] for some rolling window Δ (last 3 hours,…
• Root CA trust store — A root CA trust store is the local database of trusted certificate-authority root certificates that a TLS client consults to validate a server's certificate chain.…
• Root-entity experience resolution — Root-entity experience resolution is the specific step in Zalando's Interface Framework request lifecycle where an Experience is selected — once — for the whole request.…
• Root-leaf ML serving architecture — Root-leaf ML serving architecture is an online-inference system shape where a single root tier handles feature retrieval + preprocessing on CPU,…
• Round-trip time (RTT) — Round-trip time (RTT) is the time it takes for a signal to travel from sender to receiver and back across a network path.…
• Route leak — A route leak is a BGP advertisement propagated beyond its intended scope — formally defined in RFC 7908 as "the propagation of routing announcement(s) beyond their intended scope."
• Route Origin Validation (ROV) — Route Origin Validation is the BGP speaker-side process of checking each received route against RPKI-signed Route Origin Authorizations (ROAs).…
• Router peer — A router peer is a distinguished role in Netbird's WireGuard mesh: a peer that is permitted to accept traffic from other peers and egress it to destination resources,…
• Routing receipt — A routing receipt is the audit-grade record produced for every AI-agent task-routing decision in an OmniNode-shaped system. The receipt captures four fields, all required:
• Row-level concurrency — Row-level concurrency is the table-format property whereby two writers updating different rows of the same table do not conflict even if those rows live in the same data file.…
• Row-level lock contention — Row-level lock contention is what happens when multiple concurrent transactions need exclusive access to the same row in a row-locking database engine.…
• Row-level security — Row-level security (RLS) is the authorization mechanism that filters which rows of a dataset a given user can see, as a function of their identity.…
• Row tracking — Row tracking is the table-format property of giving every row a stable, persistent identifier that survives table-level rewrites — compaction, schema evolution,…
• Row-major vs column-major layout — Two conventions for laying out a 2D matrix in linear memory:
• RPKI (Resource Public Key Infrastructure) — RPKI is the public-key infrastructure that lets holders of Internet number resources (ASNs, IP prefixes) cryptographically attest to who is authorized to do what with them.…
• RPO / RTO (recovery point / time objectives) — The two canonical Disaster Recovery budget dimensions:
• RSS (Resident Set Size) — RSS (Resident Set Size) is the amount of non-reclaimable physical memory currently held by a process. For Postgres: "RSS is the amount of private memory allocated to a process such…
• RTNL lock contention — The Routing Netlink (RTNL) lock is Linux's single-writer mutex for network-configuration modifications (creating tap/veth devices, moving devices between namespaces,…
• Rule-set O(1) rule matching — A rule set is a data structure that maps a pair to the rules that reference it, enabling O(1)-per-query-metadata-field rule-candidate identification regardless of how…
• Rumor-mongering (gossip) — Rumor-mongering (a.k.a. the dissemination protocol) is the gossip family in which each new update is treated as a rumor that peers re-share until it is "heard enough times"…
• Run queue latency — Run queue latency is the time a Linux task spends in the scheduler's run queue — in the TASKRUNNING (runnable) state — after the kernel has decided it is ready to execute…
• Run-to-run variance — Run-to-run variance is the variability in measured latency between repeated invocations of the same benchmark on the same binary.…
• Runaway WAL growth — Runaway WAL growth is a Postgres failure mode in which the primary's WAL volume grows unboundedly because one or more logical replication slots pin WAL that the server cannot…
• Runnable plan at every step — Runnable plan at every step is a query-planner architectural invariant: every intermediate state of the optimisation pipeline — not just the final output — is a complete,…
• Runtime API vs GitOps source of truth — The runtime-API-vs-GitOps source-of-truth tension arises when a platform exposes both a declarative Git-reconciled substrate (e.g.…
• Runtime choice per workload — Runtime choice per workload is the design stance that the JavaScript runtime a function runs on is a per-workload decision, not a platform-wide commitment.…
• Runtime code sharing — Runtime code sharing is the property that two independently built and deployed JavaScript bundles can share module code — including large libraries like React — at runtime,…
• Runtime control points (compression) — A runtime control point in OpenZL is a per-frame branch point embedded in a compression Plan that reads lightweight statistics at encode time and picks a subgraph based on them.…
• Runtime dependency on SaaS provider — A runtime dependency on a SaaS provider exists whenever a system needs a third-party hosted service to be available during normal operations — not just at build, deploy,…
• Runtime environment detection — Runtime environment detection is the technique of having a single container image inspect its own runtime context at startup and adapt its behaviour — env vars, user/group ID,…
• Runtime flavor dispatch — Runtime flavor dispatch is the technique of selecting between multiple implementations of the same interface based on a *runtime flag (flavor) set once at app startup,…
• Runtime metadata attach — Runtime metadata attach is the practice of attaching dynamic context (request IDs, service endpoint names, latency buckets, user IDs, tenant IDs,…
• Runtime nullability telemetry — Runtime nullability telemetry is the practice of instrumenting the runtime — typically via a compiler plugin that emits recording shims…
• Runtime policy enforcement — Runtime policy enforcement is the control primitive of evaluating an action against policy before that action executes and returning an allow / deny / modify decision…
• Rust Waker — A Waker (std::task::Waker) is the abstract wake-up handle an async-Rust executor passes into every call to Future::poll. When a Future returns Poll::Pending (not ready yet),…
• S-curve limits — S-curve limits is the recurring framing in the Redpanda convergence of AI and data streaming series that every growth axis in frontier AI — data volume, parameter count,…
• S3 server access logs — S3 Server Access Logs (SAL) are the per-bucket access-log primitive Amazon S3 offers: every request against the bucket and its objects (GETs, PUTs, lifecycle expirations,…
• S3 signal bucket as config fanout — S3 (object storage) as the fanout substrate for configuration-management signals — a small set of well-known keys, each representing "is there new work for consumers in category…
• Safe expression language — A safe expression language is a domain-specific language designed explicitly to allow tenant-supplied code to execute in a shared process without compromising the process's safety…
• Same-day UI delivery — Same-day UI delivery is the cadence property where UI changes (screens, layouts, copy, A/B variants) can be authored, deployed,…
• Same-origin dictionary scope — Same-origin dictionary scope is the constraint — enforced by RFC 9842 — that a compression dictionary served by origin A is only usable to compress responses from origin A.…
• SAML authentication bypass — A SAML authentication bypass is any vulnerability that lets an attacker log in as an arbitrary user of a SAML-protected service-provider (SP) without the corresponding…
• Sample ratio mismatch (SRM) — Sample ratio mismatch (SRM) is a data-quality failure in A/B testing where the actual ratio of users assigned to treatment vs control differs significantly from the intended…
• Sandbox benchmarking for signal isolation — Sandbox benchmarking for signal isolation is the practice of running A/B performance benchmarks inside an ephemeral minimal-dependency container…
• Sandbox poisoning — Sandbox poisoning is the failure class in which a single request's crash leaves a shared sandbox in an undefined state, so that subsequent requests served by the same sandbox can…
• Saturation point (inference latency vs token count) — The saturation point is the token-count threshold on a specific (model, inference engine, GPU) triple at which transformer inference latency transitions from the approximately-flat…
• Software Bill of Materials (SBOM) — A Software Bill of Materials (SBOM) is a machine-readable inventory of every package and library that composes a software artifact.…
• Scalar quantization — Scalar quantization is a lossy compression technique for high-dimensional vector data that maps each float element independently to a lower-precision numeric representation…
• Scale to Zero — Scale-to-zero is a service-design property in which an application consumes no capacity and accrues no charge when it has no traffic,…
• Scaling ladder — Scaling ladder is the canonical progression of database-scaling levers, ordered from cheapest-lowest-impact to most-expensive-highest-impact.…
• Scaling latency — Scaling latency is the time between "we need more capacity" and "that capacity is actually serving traffic." For a cluster scaler, it decomposes into roughly:
• Scatter-gather query — A scatter-gather query is a query executed in a sharded system that cannot be routed to a single shard (because its predicate doesn't include the shard key or because it needs data…
• Schema as code — Schema as code is the database analogue of Infrastructure as Code (IaC): the schema of a relational database is expressed as a text file in a declarative DSL (HCL, SQL, YAML,…
• Schema-change deploy order — Schema-change deploy order is the operational rule governing which of the two deploy units (code or schema) must ship first for a given class of schema change so the production…
• Schema-change operational friction — Schema-change operational friction is Shlomi Noach's 2021 framing of the gap between the cost of executing a schema change in principle and the cost of shipping one in production.…
• Schema-change queue — A schema-change queue is a per-target serialisation primitive that orders concurrent schema-change submissions from multiple engineers and runs them in submission sequence,…
• Schema-change revertibility asymmetry — Schema-change revertibility asymmetry is the structural observation that the compensating DDL for an ADD COLUMN and the compensating DDL for a DROP COLUMN are not symmetric…
• Schema-Constrained LLM Output — Schema-Constrained LLM Output is the technique of binding an LLM call to a target output schema (typically JSON) so the model's response must parse into the declared shape…
• Schema dependency graph — A schema dependency graph is the directed graph whose nodes are schema entities (tables, views, columns, indexes, foreign-key constraints, check constraints,…
• Schema-diff equivalence class — A schema-diff equivalence class is a connected component of the dependency graph formed by a set of schema-diff statements.…
• Schema disagreement — Schema disagreement is a state where different nodes of a distributed database hold different versions of the schema — as advertised via the schema version that gossips…
• Schema discoverability gap in errors — The schema discoverability gap in errors is the structural gap in GraphQL where the schema describes response.data but says nothing about the shape or vocabulary…
• Schema discovery vs data access separation — A data-platform governance affordance in which users can see what tables exist (their names, possibly some lineage and ownership metadata) but cannot see column-level information…
• Schema evolution — Schema evolution is the problem of changing the structure of data (records, tables, messages) over time while old and new versions coexist in the system — in flight on a queue,…
• Schema linter enforcement — Schema linter enforcement is the discipline of running a build-time static-analysis tool over an API schema and failing the build when the schema violates the organisation's…
• Schema registry context — A schema registry context is a namespace boundary inside a schema registry that isolates schemas into logical groups sharing the same physical registry instance.…
• Schema registry — A schema registry is a centralized, versioned store of data contracts — typically event / message / record schemas — used as the single source of truth for the shape, type,…
• Schema revert — Schema revert is the operational primitive of undoing a deployed schema change while preserving the application-level writes that landed after the change was deployed.…
• Schema routing rules — Schema routing rules are a Vitess feature (docs) that let operators declare, at the keyspace / table / workflow level, which backing storage a given query should be routed to.…
• Schema three-way merge — Schema three-way merge is the schema-flavoured analog of Git's three-way merge: given a base schema (main) and two developer branches (branch1,…
• Scream test for deletion — A scream test is a reversible pre-delete state that makes a resource behave (from the outside) as if it is already deleted,…
• Seamless signal switching — Seamless signal switching is viewer-imperceptible failover between hot-standby redundant broadcast streams. Gaps on one contribution leg are concealed by another leg at the packet…
• Seasonality (daily / weekly) — Seasonality is the property of a time series in which values recur with approximate periodicity at a named timescale — daily, weekly, monthly, yearly.…
• seccomp — seccomp (short for secure computing mode) is a Linux kernel feature that restricts the set of system calls a process is allowed to make.…
• Secondary index — A secondary index is a database index on a non-primary-key column (or column set). In a system with a clustered index like MySQL's InnoDB,…
• Seconds-scale GPU cluster boot — The property of a compute platform where a multi-node cluster of GPU-equipped machines, defined by a Docker image, can be booted in seconds rather than minutes*.…
• Secure aggregation — Secure aggregation is the class of cryptographic protocols that lets a server compute a sum (or other aggregate) of values from many clients without learning any individual…
• Security group profile — A security group profile is a layered, pre-approved set of AWS security-group rules that reflect an organisation's environment tiers (e.g.…
• Security model is the data model — A design principle for AI data agents: the agent has no authority of its own — it runs as the calling user, inheriting exactly that user's permissions on the underlying data.…
• Segment-level Root-cause Diagnosis — Segment-level root-cause diagnosis is the observability discipline of aggregating noisy per-item scores (e.g. per-query / per-result relevance) up to a segment…
• Segmented Changelog (algorithm) — Segmented Changelog is an algorithmic representation of a VCS commit graph that partitions commits into maximal linear segments at merge points and stores the inter-segment merge…
• Segmented partner data access tiering — A segmented partner data access tiering is the design recognition that a B2B data provider's partner base does not have uniform access needs,…
• SELECT FOR UPDATE SKIP LOCKED — SELECT ... FOR UPDATE SKIP LOCKED is a row-locking clause that acquires a row-level exclusive lock on the matching rows and skips over any rows another transaction has already…
• SELECT-only index-telemetry caveat — The invariant: PlanetScale Insights' per-query-pattern index-usage telemetry — surfaced via the Indexes tab, the index:table.index search predicate,…
• Selection metadata (Zalando Experience) — Selection metadata is the declarative predicate a Zalando Experience carries to tell the domain backend when this Experience is eligible.…
• Selective FP8 quantization — Selective FP8 quantization is a post-training quantisation strategy that applies FP8 precision only to model layers with high precision-loss tolerance,…
• Selectively reliable video transport — Selectively reliable video transport is the research direction of making QUIC's reliability model per-frame selective: not every video frame must be reliably delivered.…
• Self-approval bias — Self-approval bias is the failure mode where a single LLM instance that both generates an output and evaluates it systematically over-approves its own work.…
• Self-authored extension — A self-authored extension is a tool that an AI agent writes for itself at runtime — as source code, bundled + loaded into a sandbox with an explicit permission declaration…
• Self-censoring forecast — A self-censoring forecast is a predictor that scores its own recent accuracy and only emits a prediction when that accuracy has been high.…
• Self-healing job queue — A self-healing job queue is an async-job system architected so that losing the queue's contents — whether from Redis data loss, operator mistake, corruption,…
• Self-inflicted DoS — A self-inflicted denial of service is an outage where the saturating traffic originates from an internal, trusted client — not from an external attacker — and is typically valid,…
• Self-invalidating forecast — A self-invalidating forecast is the hazard class where a control loop forecasts a metric whose future value depends on the control action being taken on the forecast.…
• Self-serve generative AI — Self-serve generative AI is the UX posture where an organisation's internal generative-AI platform is usable by anyone in the company regardless of technical background…
• Self-service infrastructure — Self-service infrastructure is the platform-engineering property where product developers can provision and modify their own infrastructure resources — compute, capacity,…
• Self-signed certificate anti-pattern (for managed services) — When a client cannot validate a managed-service endpoint's TLS certificate (because the client is missing a public-CA trust store), two tempting "fixes" circulate in forums:
• Semantic enterprise context — Semantic enterprise context is the rich, structured information implicit in the relationships between an organisation's data assets — table schemas, dashboard definitions,…
• Semantic ID — A Semantic ID is a discrete-token identifier for a recommendable item, encoded as a short sequence of codewords from a learned hierarchical codebook,…
• Semantic interoperability — Semantic interoperability is the property that two systems can exchange data and preserve its meaning — not just its shape.…
• Semantic layer of business concepts — A semantic layer of business concepts is an intermediate layer — between source-system schemas and a target schema (such as the logical data model of a golden record)…
• Semantic merge conflict — A semantic merge conflict is the failure mode where two pull requests each pass branch-level CI cleanly in isolation but, when both merge into the same target branch concurrently,…
• Semantic Role Labeling (SRL) — Semantic Role Labeling (SRL) in the e-commerce query-understanding context is the task of extracting structured slots — product, brand, attribute, size, quantity,…
• Semantic schema diff — A semantic schema diff is the set of SQL DDL statements (ALTER TABLE, CREATE TABLE, DROP TABLE, ALTER VIEW, etc.) that transform one database schema into another.…
• Semantic search over agent memory — Semantic search over agent memory is the retrieval model where an AI agent's persistent memory store is accessed via natural-language queries resolved against vector similarity…
• Semantic search retrieval — Semantic search retrieval is meaning-based search that retrieves related content based on conceptual similarity rather than exact keywords or explicit references.…
• Semi-sync timeout fallback — Semi-sync timeout fallback is the specific MySQL behaviour in which a primary, having waited rplsemisyncmastertimeout milliseconds for a sufficient number of semi-sync replicas…
• sendBeacon telemetry transport — sendBeacon() (navigator.sendBeacon) is a browser API designed for sending small amounts of data to a server asynchronously and reliably, at a low priority,…
• Send what you use — Send what you use is the design principle that a producer of an RPC payload should include only the fields the receiver will actually consume,…
• Sender reputation — Sender reputation is an ISP-tracked score that estimates how trustworthy a sender (identified by IP address, sending domain, or both) is.…
• Sensitive data exposure — Sensitive data exposure is a security vulnerability in which confidential or protected information is unintentionally made accessible to parties that should not have access,…
• Sensitive field logging redaction — Sensitive field logging redaction is the discipline of ensuring that designated fields or arguments — typically PII, credentials,…
• Separate binlog disk — A separate binlog disk is a MySQL operational posture in which the binary log (binlog) — the sequential append-only log of every committed transaction…
• Separate vs. monolithic data models — The fundamental structural choice when a data warehouse must support multiple product lines or business domains.
• Separation of concerns — Separation of concerns is the practice of organising a system so that each component has a single, well-defined responsibility,…
• Separation of duties (data governance) — Separation of duties in data governance is the structural distribution of governance responsibilities across specialised roles,…
• Sequence quality dimensions — Sequence quality in a multi-tenant ML data platform is multi-dimensional — not a single SLO but a vector of four orthogonal properties that consumers (training pipelines,…
• Sequential vs random I/O on SSD/EBS — The same byte-level I/O demand has very different IOPS cost depending on whether the reads/writes are sequential (contiguous in the logical block address space) or random…
• Serializable isolation — Serializable is the strongest ANSI-SQL isolation level. It guarantees that concurrent transactions yield the same result as some serial execution of the same transactions run one…
• Serialization tax in proxy path — Serialization tax in proxy path is the latency cost imposed on every request by an in-path proxy that must deserialize + re-serialize the payload to make a routing decision.…
• Serialized per-shard updates — Serialized per-shard updates is Temporal's consistency discipline: all state mutations for workflows that hash to the same shard are applied to the persistence layer sequentially,…
• Server-driven UI — Server-driven UI (SDUI) is the architectural choice to have the server describe, per request, the UI the client should render — the structure (views, layouts, sections),…
• Server-owned schema versioning — Server-owned schema versioning is the discipline of keeping all component-version negotiation logic on the server in a server-driven UI system, so that:
• Server-side request forgery (SSRF) — Server-side request forgery (SSRF) is the attack class where an attacker coerces a server to make an unintended internal HTTP request on the attacker's behalf.…
• Server-Side Sandboxing — Server-side sandboxing (also called workload isolation) is the practice of running a given piece of server-side code inside a restricted execution environment whose capabilities…
• Serverless billing definition — Simon Willison's crystallization of "serverless":
• Serverless Compute — Serverless compute is a model in which the provider runs customer code on demand on managed, shared infrastructure, scaling it from zero to arbitrary concurrency without…
• Serverless TCP socket restriction — Serverless TCP socket restriction is the platform constraint that serverless / FaaS / edge compute runtimes forbid user code from opening raw TCP sockets,…
• Serverless workload churn cardinality — Serverless and ephemeral-workload platforms multiply metric cardinality faster than steady-state fleet growth would suggest.…
• Service Control Policy (SCP) — A Service Control Policy (SCP) is an AWS Organizations policy that defines the maximum set of permissions any principal in an AWS account (or OU) can use…
• Service coupling — Service coupling is the degree to which services depend on each others' implementation, availability, and behavior. Tight coupling produces short,…
• Service dependency graph — A service dependency graph is a runtime-derived map of which services in a distributed system actually call which other services,…
• Service-enforced deletion protection — Service-enforced deletion protection is the property where a storage system *prevents deletion or retention-shortening of stored artefacts by any principal…
• Service group — A service group is a clustered collection of workloads (pods, containers, processes, hosts) that share a common identity in observability data and are addressable as a single…
• Service Level Indicator (SLI) — A Service Level Indicator (SLI) is the measurement a Service Level Objective is defined over — the actual metric you collect.…
• Service Level Objective (SLO) — A Service Level Objective (SLO) is a target reliability level for a service, expressed as a threshold over a Service Level Indicator (SLI) — e.g.…
• Service-oriented architecture (SOA) — Service-oriented architecture (SOA) is the design pattern of decomposing a system into small services, each dedicated to a specific well-encapsulated domain area (e.g.…
• Service tier classification — Service tier classification is the discipline of labelling services (or operations) by criticality tier — typically Tier-1 (revenue-critical / user-hot-path),…
• Service topology — Service topology is a configuration abstraction answering "at which locations is this service's IP address allowed to be reachable from the Internet?".…
• Session-affinity prompt caching (x-session-affinity) — Session-affinity prompt caching is the LLM-serving pattern of routing subsequent turns of a session back to the same replica / region that served the previous turn,…
• Session-cookie read-your-writes window — A session-cookie read-your-writes window is the mechanism of using a short-lived per-user cookie (or equivalent session-state marker) to pin that user's reads to the primary…
• Session identity evaluation (for views and functions) — Session identity evaluation is the rule that policies inside a SQL view or function evaluate against the identity of the user running the outer query…
• Session-level system setting — A session-level system setting is a MySQL configuration variable whose value is scoped to a single connection — set once,…
• Session-token-embedded routing context — Session-token-embedded routing context is the design pattern in which a session token (the bearer credential proving a logged- in user) also carries tenant/shard routing…
• SET_VAR optimizer hint — SETVAR is a MySQL 8.0 optimizer hint that applies a session system variable for the duration of a single query without modifying the underlying connection state.…
• Severity-filtered violation reporting — Severity-filtered violation reporting is the practice of surfacing only the highest-severity bucket of violations when a new automated-check class is introduced,…
• SHACL (Shapes Constraint Language) — SHACL — Shapes Constraint Language — is the W3C standard for expressing validation constraints over RDF graphs. A SHACL shape describes a class of target nodes and the constraints…
• Shadow cluster — A shadow cluster is a parallel cluster running a candidate release of a system, which receives a mirror of production traffic so that its behaviour can be compared against…
• Shadow job (pre-production) — A shadow job is a job that runs in a pre-production environment, consumes the same source as a production job, and writes its output to a separate shadow table…
• Shadow-mode alert validation — Shadow-mode alert validation is the discipline of deploying a new alerting rule (or a new symptom source like an e2e test probe) in a lower-severity,…
• Shadow-mode metric collection — Shadow-mode metric collection is the practice of running a candidate integration (new API, new downstream, new code path) in parallel to the existing production path…
• Shadow table — A shadow table is a second, hidden table created in the same database as the production table it mirrors, carrying either the same schema (for zero-downtime moves / reshards)…
• Shadow WAL (Litestream legacy mechanism) — The shadow WAL is the mechanism the original 2020 Litestream used to get a coherent, durable copy of a SQLite WAL out of a live database without modifying the application or SQLite…
• Shallow-run detection — A health signal that flags any hunter agent finishing with zero findings as "shallow" and immediately requeues it for a new run…
• Shard allocation awareness — Shard allocation awareness is Elasticsearch's mechanism for telling the cluster about a physical distribution attribute of each node — typically availability zone, rack,…
• Shard key cardinality — Shard key cardinality is the number of distinct values a candidate shard-key column takes across the dataset. A high-cardinality column (e.g.…
• Shard-key as database_id — databaseid as shard key is the canonical shape for a telemetry / multi-tenant store where each customer database is the unit of isolation and no workflow joins across customer…
• Shard key volatility — Shard key volatility is how often the value in the shard-key column changes for a given row. An immutable shard key (userid, orgid,…
• Shard key — A shard key is the column (or composite) whose value selects which physical shard a row lives on under horizontal sharding.…
• Shard-parallel backup — Shard-parallel backup is the architectural property of a horizontally-sharded database where backup wall-clock time scales down as N × per-server-bandwidth where N is the shard…
• Sharded failure-domain isolation — Sharded failure-domain isolation is the architectural property that a horizontally-sharded database's outage blast radius is bounded to the customers assigned to the failing shard…
• Shared-context fan-out — Shared-context fan-out is the pattern of writing large common context (merge-request metadata, previous review findings, diff summaries) to a shared file on disk and pointing…
• Shared-dictionary compression — Shared-dictionary compression is an HTTP-level compression technique in which server and client share a reference payload (the "dictionary") that both sides know about,…
• Shared-everything architecture — A shared-everything architecture stores all data in a single shared storage system accessible to every compute node in the cluster.…
• Shared fault domain — A shared fault domain is any infrastructure element — a Kubernetes node, availability zone, rack, network switch, control plane, cloud region, CI pipeline,…
• Shared lock vs exclusive lock — Two of the foundational row-level lock kinds in InnoDB (and most row-locking RDBMS engines): shared locks (S) and exclusive locks (X).…
• Shared mobile foundations — Shared mobile foundations are the cross-cutting libraries and platform services (identity, real-time monitoring, performance measurement, networking, error reporting,…
• Shared-nothing architecture — A shared-nothing architecture gives each node in the cluster its own private storage; no two nodes share a physical storage resource.…
• Shared-nothing PHP request model — The shared-nothing PHP request model is PHP's foundational language-runtime design decision that every HTTP request executes in a fresh PHP process image,…
• Shared Payment Token (SPT) — A Shared Payment Token (SPT) is a payment primitive built for agentic commerce that lets an AI agent initiate a payment with a human buyer's permission and preferred payment…
• Shared Responsibility Model — The Shared Responsibility Model is AWS's contract-level framing for which party — AWS or the customer — is responsible for what parts of a running workload.…
• Shared singleton dependency — A shared singleton dependency is a library that must have exactly one module instance on the page even when multiple independently built bundles each declare a dependency on it.…
• Shared test environment contention — Shared test environment contention is the failure mode where a single test/staging environment is the serialisation point for multiple in-flight changes…
• Shift-left privacy — Shift-left privacy is the engineering stance of moving privacy enforcement earlier in the data-lifecycle — from after-the-fact audits and periodic reviews into the developer…
• Shift-left validation — Shift-left validation is the engineering stance of moving a correctness check earlier in a change's lifecycle — from runtime to deploy time, from deploy time to apply time,…
• Shim layer — A shim layer is a thin proxy library interposed between application code and one or more underlying implementations, exposing a single unified API and dispatching each call…
• Shopping conversion candidate generation — Shopping conversion candidate generation is the retrieval-stage ads-ML primitive that narrows a large shopping-Pin (or shopping-product) catalog to a rank-tractable set…
• Short-lived credential auth — Short-lived credential auth is the security property that a workload's authorization is carried by credentials with minutes- scale lifetime, generated dynamically per session,…
• Short-term vs long-term engagement — Short-term engagement — immediate actions on a single impression or session (clicks, saves, watch-time, purchases). Long-term engagement — session length, revisit likelihood,…
• Shot-level embedding — A shot-level embedding is a vector representation computed per shot of a piece of long-form video content — where a "shot" is the atomic temporal unit of video produced by shot-…
• Shuffle sharding — Shuffle sharding is a tenant-isolation technique that gives each tenant in a shared cluster a single-tenant experience by assigning them a randomly-chosen subset of backend nodes…
• Sidekiq unique jobs — Sidekiq Enterprise unique jobs is a paid-feature mechanism that prevents duplicate jobs from being enqueued within a configurable time window.…
• Signal-based step dependency — A signal-based step dependency is a data-condition-driven way to unblock a workflow step: instead of waiting on upstream step completion (the usual DAG dependency shape),…
• Signal-driven Chef trigger — Signal-driven Chef trigger is the architectural choice of running chef-client only when a new cookbook version is available (via an out-of-band signal),…
• Signal-handler fork inheritance — Signal-handler fork inheritance is the set of POSIX rules that govern what happens to installed signal handlers when a process calls fork(2). The rules are simple on paper:
• Signal-to-noise in AI vulnerability triage — Signal-to-noise in AI vulnerability triage is the operational problem that LLM-driven vulnerability scanners produce many more hedged, speculative findings than they do real bugs,…
• Signed embed URL — A signed embed URL is a short-lived, cryptographically-signed URL that carries the authorization context of one specific viewer.…
• Signing key rotation lifecycle — The signing-key rotation lifecycle is the ordered discipline for replacing a cryptographic signing key without invalidating tokens in flight and without requiring any coordination…
• Silent hang (LLM inference server) — A silent hang is a failure mode in LLM inference engines where the server stops responding to requests but does not surface an error — no exception logged,…
• Silent wiring failure — Silent wiring failure is the bug class in which a producer publishes to topic A and a consumer subscribes to topic A' where A and A' are both well-formed and both accepted…
• SIMD vectorization — SIMD — Single Instruction, Multiple Data — is a CPU instruction class where one instruction performs the same arithmetic operation on multiple data elements in parallel,…
• Similarity-depth correlation — Similarity-depth correlation is an intrinsic-evaluation metric for a hierarchical-codebook Semantic ID system: the Spearman correlation between the continuous embedding similarity…
• Similarity-tier retrieval — Similarity-tier retrieval names the product + eval constraint that a search feature must return high-quality results at multiple similarity levels simultaneously — exact matches,…
• Simplicity vs. Velocity — A first-class engineering concept in S3's 2025 retrospective: there is a real, two-way tension between shipping a feature quickly and shipping a feature that is simple enough…
• Single-agent coverage failure on large repos — Single-agent coverage failure is the empirical observation that a single LLM-coding-agent session, no matter how capable the model,…
• Single-controller RL orchestration — Single-controller RL orchestration is an execution model for on-policy RL post-training in which the driver node stops being a passive launcher of identical workers and becomes…
• Single-endpoint abstraction — Single-endpoint abstraction is the architectural principle of giving clients one stable address for a heterogeneous or dynamically-changing backend fleet,…
• Single pipeline train-and-infer — Single pipeline train-and-infer is an ML-pipeline architectural pattern where a single compute job owns both model training and batch inference in one pass,…
• Single-platform application architecture — A single-platform application architecture runs the application, the operational database, the analytical store, the ML/training substrate,…
• Single-shard throughput ceiling — The single-shard throughput ceiling is the per-shard operation rate a sharded system can sustain before saturating that one shard, independent of the rest of the fleet.…
• Single-sink telemetry architecture — Single-sink telemetry architecture is the structural posture of routing observability traffic (traces / logs / metrics) directly from instrumented clients to a single durable…
• Single-source interface spec — A single-source interface spec is a language-agnostic description of a component's (or API's, or message's) public interface — field names, field types, default values,…
• Single-use virtual card for agents — A single-use virtual card for agents is a payment credential issued by a wallet / issuer to an AI agent for one authorised transaction only, carrying scope-fields (amount,…
• Single-vs-multiple-Organizations decision — The single-vs-multiple-Organizations decision is the architectural choice enterprises face when adopting AWS Organizations: run one organization that contains all AWS accounts…
• Single-writer assumption — An architectural posture in which a system assumes exactly one writer is active against a given piece of state at any moment,…
• Singleton container pattern — A singleton container is a Docker container declared once per JVM test process, started in a static initialiser, held in a static field,…
• Singleton workload — A singleton workload is a service that runs as a single replica — one pod, one instance, no horizontal redundancy. It may be stateful (a leader-elected controller,…
• Sitemap — A sitemap is an XML file listing every URL on a site, plus per-URL metadata (last-modified, change frequency, priority). The standard is sitemaps.org (2005).…
• Sketch as MySQL binary column — Sketch as MySQL binary column is the pattern of storing a quantile-sketch data structure (DDSketch, t-digest, HDRHistogram) as a BLOB column in a MySQL row,…
• Sketching feature store — A sketching feature store is a machine-learning feature store whose backing storage is a probabilistic data structure — most commonly a Bloom filter…
• Skewed column selectivity — Skewed column selectivity is the condition where a column's selectivity computed on average (across the whole table) is a poor predictor of selectivity for specific predicate…
• SKUs as time-series unit — In e-commerce demand-forecasting systems, each SKU (Stock-Keeping Unit) is treated as an independent time-series unit: the forecast is computed per SKU,…
• Sliding Spectrum Decomposition (SSD) — Sliding Spectrum Decomposition (SSD) is a position-adaptive diversification method for recommendation systems. Views a candidate feed as a mixture of latent spectra (topics,…
• Sliding-window epoch tracking — Sliding-window epoch tracking is the per-shard mechanism that publishes a monotonically non-decreasing safe-to-GC watermark M(p) for use in epoch-based distributed garbage…
• Sliding-window training — Sliding-window training is a time-series training strategy where the training dataset is a fixed-length window that slides forward in time with each retraining cycle.…
• Slot-vs-offset position tracking — Slot-vs-offset position tracking is the structural problem that Postgres logical-replication CDC consumers (notably Debezium) track the same stream position in two independent…
• Slow is failure — Slow is failure is the operational premise that for customer-facing real-time workloads — OLTP databases, API backends, interactive request/response…
• Slowly-Changing Dimension (SCD) — Slowly-Changing Dimension (SCD) is the dimensional-modeling pattern for dimension tables whose attributes change over time — customer address, product category,…
• Small-batch NVMe write amplification — Small-batch NVMe write amplification is the storage-substrate cost that a streaming broker incurs when producer batches are smaller than the NVMe SSD's 4 KB page-alignment unit.…
• Small file problem on object storage — The small file problem is the pathology of a streaming-to-lakehouse pipeline producing many small object-store files (Parquet / ORC / Avro) instead of fewer, well-sized ones.…
• Small map as sorted vector — Represent an associative container whose key count is empirically small and bounded as a flat sorted Vec<(key, value)> (a.k.a.…
• Smart chunking — Smart chunking is a document-structure-preserving chunking strategy for LLM-based document intelligence — chunks are constructed to retain hierarchy and one-to-many relationships…
• SMART on FHIR — SMART on FHIR ("Substitutable Medical Applications and Reusable Technologies on FHIR") is the auth + data-exchange profile that lets third-party clinical apps and patient-facing…
• Snapshot-diff inference CDC — Snapshot-diff inference CDC is the class of CDC ingestion where the upstream source does not emit a native change data feed;…
• Snapshot Isolation — Snapshot Isolation (SI) is a transaction-isolation model in which each transaction reads from a consistent snapshot of the database as of its start time: it sees all writes…
• Snapshot repository as agent corpus — Snapshot repository as agent corpus names the architectural choice of giving an agent a dedicated versioned repository as its knowledge-base view,…
• Snowflake ID — A Snowflake ID is a 64-bit time-ordered identifier originally designed by Twitter (2010) for generating unique post IDs across a sharded fleet without coordination.…
• Snowflake MATCH_RECOGNIZE sessionization — MATCHRECOGNIZE is Snowflake's SQL clause for row-pattern matching over an ordered event stream. Given a partitioning key (e.g. ip), an ordering key (e.g.…
• Snowpipe Streaming channel — A Snowpipe Streaming channel is the per-table parallelism unit of Snowflake's low-latency row-level ingest API (Snowpipe Streaming).…
• Social graph (objects and edges) — The social graph as objects + edges is the canonical relational encoding of a social-media-style entity-and-relation model, where:
• Soft-delete vs hard-delete — A canonical application-design trade-off for handling row deletion in production databases:
• Soft leader election — Soft leader election designates one pod as the coordinator for a given key by routing affinity — if all requests for a key land on the same pod,…
• Soft-spacing penalty — A soft-spacing penalty is a distance-weighted score reduction applied to a candidate item when nearby already-placed items share a sensitive content class.…
• Software-induced carrier outage — A software-induced carrier outage is a carrier-scale voice and data service disruption caused by an internal software defect — not physical damage, not a weather event,…
• Somewhat Relevant Evaluation Category — A "somewhat relevant" evaluation category is the distinctive middle-tier rubric label used in LLM-as-judge grading of semantic search results,…
• Source and target domain — In transfer learning and domain-adaptive learning, source domain and target domain name the two settings between which knowledge is transferred:
• Source code as agent feedback loop — Source code as agent feedback loop is the observation that merged source code becomes an LLM-based coding agent's implicit long-term memory,…
• Source-map composition — A source map is a file that maps locations in compiled output (JS, WASM, minified bundle) back to locations in the source the developer wrote.…
• Source-of-truth disambiguation — Source-of-truth disambiguation is the data-agent reasoning capability of determining the most authoritative information when multiple sources (table metadata, company documents,…
• Source-of-truth hydration — Source-of-truth hydration is the ingestion-side primitive of calling back to the source system on each event to fetch the complete current state,…
• Space-based compute — Space-based compute is the architectural choice to deploy the compute substrate in orbit rather than terrestrially. Drivers:
• SPANN index — A SPANN (Space-Partitioned Approximate Nearest Neighbors) index is a hybrid tree + graph ANN index from Microsoft Research (NeurIPS 2021) designed specifically for larger-than-RAM…
• Spanner change stream — Spanner change streams are Google Cloud Spanner's native CDC mechanism: a named change stream is declared against a database, table set, or column set,…
• Spark ETL Feature DAG — Spark ETL Feature DAG is an architectural model for structuring Spark-based ETL pipelines as a directed acyclic graph of features — self-contained,…
• Sparse checkout — Sparse checkout is a VCS feature that materializes only a declared subset of the repository's paths to the working copy on disk.…
• Sparse lexical retrieval — Sparse lexical retrieval is the class of information-retrieval approaches that match a query to documents by exact or closely-matched terms stored in an inverted index,…
• Sparse profile — A sparse profile is a named, checked-in description of which subset of the repository a particular product / team / role needs in their working copy.…
• Sparse Vector — A sparse vector is a vector representation where most components are zero, stored and queried via a coordinate format ({index: value} pairs) rather than a dense array.…
• Spatial locality (prefetching) — Spatial locality is the property that data stored near (physically adjacent, sequentially indexed, temporally contiguous) a just-accessed item is more likely to be accessed next.…
• Spec-compliant optimization — Spec-compliant optimization is the design discipline of removing allocations, scheduling hops, or work from an implementation within the observability constraints the spec imposes…
• Specialized knowledge search — Specialized knowledge search is an asset-discovery technique used by Databricks Genie for finding the relevant tables, dashboards, notebooks,…
• Specification-driven development — Specification-driven development is a workflow where the specification is a first-class, authored, maintained artifact — produced early, visible to customers,…
• Speculative decoding — Speculative decoding is an LLM-inference latency-optimization technique: a small, fast drafter model proposes the next N tokens autoregressively, and a large,…
• Speech recognition (ASR) — Automatic Speech Recognition (ASR) is the ML primitive that converts an audio signal of human speech into a text transcript.…
• Speed-Accuracy Trade-off (Real-Time Decisions) — Structural property of real-time decision systems where the data supporting a decision accumulates over time, so the decision's accuracy improves monotonically with the time…
• SPFresh index — An SPFresh index is a SPANN ANN index extended with concurrent background maintenance operations — split, merge, rebalance,…
• SPICE in-memory caching — SPICE — Super-fast, Parallel, In-memory Calculation Engine — is Amazon QuickSight's columnar in-memory cache. Datasets loaded into SPICE are held in a compressed columnar…
• Spiky traffic — Spiky traffic is the pattern where request arrivals have high variance on short timescales — bursts arriving within seconds followed by troughs,…
• Splay — randomised run jitter — Splay (named by Chef) is a per-node randomised delay inserted between "I received a new config signal" and "I start the config-management run".…
• Split-band audio coding — Split-band (aka subband) coding splits the input audio spectrum into two or more frequency bands, encodes each band separately, and recombines at the decoder.…
• Split-brain — Split-brain is the failure mode in which two (or more) nodes each believe they have authoritative ownership of the same resource — typically a key, partition, or leadership role.…
• Split-instruction race window — A split-instruction race window is the class of correctness bugs that arises when a compiler / assembler expresses a logically single update to runtime-observable state as multiple…
• SPMD execution model — SPMD (Single Program, Multiple Data) is a distributed-execution model in which every worker runs the same step function on a different shard of the data,…
• Spurious-wakeup busy-loop — A spurious-wakeup busy-loop is the pathology where an async-Rust (or more generally, any poll-driven) state machine signals readiness to its executor without actually having…
• SQL efficiency ratio — SQL efficiency = db.SQL.tupfetched / db.SQL.tupreturned — the ratio of rows consumed by clients to rows returned from storage, measured by the Postgres stats collector.…
• SQL on FHIR ViewDefinitions — SQL on FHIR ViewDefinitions is "a new HL7 standard that flattens nested FHIR resources into tabular views for analytics." (Source:…
• SQL-to-text transformation — SQL-to-text transformation is the LLM step that converts a raw SQL query into a structured natural-language description of the business question the query was designed to answer,…
• SQLancer logic bug — A SQLancer logic bug is a class of DBMS bug in which a query returns a result set that is logically incorrect — e.g. the DBMS omits rows that should have been returned,…
• SQLCommenter query tagging — SQLCommenter is a Google-originated (google.github.io/sqlcommenter) open standard for appending key-value metadata to SQL queries as a SQL comment.…
• SQLite transaction for atomic resource claim — SQLite transaction for atomic resource claim is the use of a single transactional UPDATE...WHERE...RETURNING (with the row- selection logic pushed into the SQL) to atomically…
• SQLite Virtual Filesystem (VFS) — A SQLite Virtual Filesystem (VFS) is a plugin layer inside SQLite that abstracts the operations SQLite performs on the underlying OS — open, read, write, lock, sync, size,…
• SQLSTATE 53000 Traffic Control error — SQLSTATE 53000 is the Postgres error class ([insufficientresources][pg-errcodes]) that PlanetScale Traffic Control raises when a query in Enforce mode exceeds its budget.…
• SRE curriculum — An SRE curriculum is a structured, async-consumable body of training material on reliability practices — typically video + quiz per topic,…
• SRE KPI portfolio — An SRE KPI portfolio is the minimum measurable-outcome set an SRE function adopts to steer its own work. Without a portfolio,…
• SRE organizational evolution — SRE organizational evolution names the recurring three- phase pattern by which engineering organizations scale Site Reliability Engineering practice from a handful of on-call teams…
• SRE program — An SRE Program is a time-bounded coordination structure spanning multiple SRE teams in different reporting chains, that unites them under a shared goal-set, roadmap,…
• SSD garbage collection — SSD garbage collection is the firmware-level background process that reclaims dirty pages on an SSD, which — because NAND flash can only be erased at block granularity,…
• SSD parallelism via targets — SSD parallelism via targets is the hardware-level parallelism an SSD gets from having multiple independent NAND flash targets (dies / planes),…
• SSH job execution anti-pattern — The SSH job execution anti-pattern is the use of direct SSH into compute clusters as the submission and lifecycle substrate for batch jobs…
• SSL handshake as per-request tax — SSL handshake as per-request tax is the condition where a request-processing model opens a fresh TLS connection per request — rather than reusing a persistent connection…
• SSO authentication (OpenID Connect) — Single sign-on (SSO) is the pattern where a user authenticates once against an identity provider (IdP) and then uses the resulting token to prove their identity to many downstream…
• Stability as a system property — Stability as a system property is the architectural inversion that moves responsibility for keeping a distributed system stable from the user (who tunes cluster sizes, retries,…
• Stability threshold rollout tiers — Stability threshold rollout tiers is the practice, for a staged production rollout, of pre-committing to explicit stability thresholds and the response actions at each…
• Stack tag enrichment — Stack tag enrichment is the practice of annotating and filtering profile call stacks at query time with a small DSL that can:
• Stack-trace sampling profiling — Stack-trace sampling profiling is a production profiling technique in which a profiler periodically (e.g., 100 Hz) takes a snapshot of the call stack on each running thread,…
• Stack unwinding — Stack unwinding is the runtime operation of walking backwards through a thread's chain of function calls, starting from the current stack pointer and following saved frame /…
• Stacked immutable databases — Stacked immutable databases is a mechanism for representing changes on top of a base database without destructively modifying the base.…
• Stage and Commit — Stage and commit is a synchronisation pattern borrowed from version control (git) and applied to storage by S3 Files (2026).…
• Staged-then-sealed migration — A staged-then-sealed migration is an online schema change that deliberately holds its cut-over indefinitely: the migration backfills the shadow table,…
• Staging pipeline — A staging pipeline is a parallel data-pipeline configuration that runs alongside the production pipeline, consuming production data but executing code under test,…
• Stale-quote rejection — A client-request design in which the client-observed state (typically a price, option, or version value) is submitted back to the server as part of the write request,…
• Stale-while-revalidate cache — Stale-while-revalidate (SWR) is a caching semantic where a cache is permitted to serve a stale entry immediately while it revalidates (or re-fetches) in the background.…
• Staleness-vs-load trade-off — The staleness-vs-load trade-off is the explicit architectural choice to accept a bounded staleness window on cached data in exchange for bounded load on the backing system.…
• Standalone vs Streams mode — Redpanda Connect ships with two deployment modes optimised for different operational profiles:
• Star schema — A star schema is a dimensional data-warehousing schema design (Kimball, 1996) consisting of:
• State batching — State batching is the rendering-pipeline property of collapsing multiple setState calls within the same logical update into a single render pass.…
• State in database, not queue — The authoritative state-of-record for pending work lives in the application database, not in the job queue. The queued job is a derivative…
• State-transition logging — State-transition logging is the reliability property of emitting a log record for every transition of a deterministic state machine — not every operation, not every event,…
• Stateful incremental VM build — A VM lifecycle where the running-VM filesystem is the unit of build, mutated iteratively after boot — packages installed, source files edited,…
• Stateful stream processing — Stateful stream processing is a stream computation model where operators maintain mutable state across events — enabling aggregations, joins, sessionization, pattern detection,…
• StatefulSet for databases — A StatefulSet is the Kubernetes workload resource designed for stateful applications — applications whose identity and persistent data must survive pod restarts and reschedules.…
• StatefulSet highest-ordinal scale-in — A Kubernetes StatefulSet scales in by removing the pod with the highest ordinal — if the StatefulSet has pods statefulset-0, statefulset-1, …, statefulset-N,…
• Stateless agent compute — The principle of treating LLMs as stateless compute engines where all persistent state lives outside the model's context window in an external store (database, filesystem, queue).
• Stateless Compute — Stateless compute is a contract in which the execution environment keeps no durable state across invocations — any persistence lives in a separate managed store.…
• Stateless NAT via eBPF — Stateless NAT via eBPF replaces Linux's stateful iptables + conntrack NAT with an eBPF program that mangles packet headers based on a predetermined mapping,…
• Stateless processing — Stateless processing is a service-level discipline where, after a session completes, the service retains no access to the data it processed — there is no disk-write path,…
• Stateless rendering (Googlebot) — Stateless rendering is the property of Google's Web Rendering Service that every page render happens in a fresh browser session with no cookies, no session state, no localStorage,…
• Static/dynamic prompt partitioning — Static/dynamic prompt partitioning is the prompt-layout primitive that splits a prompt into two contiguous regions — a static prefix that is byte-identical across a batch…
• Static hash partitioning — Static hash partitioning assigns keys to nodes via a fixed modulo: node_id = hash(key) mod N where N is the total node count.…
• Static IP allowlisting — Static IP allowlisting is the common enterprise-integration requirement that an external caller can only reach a target at a fixed, stable,…
• Static sharding — Static sharding pins application keys to backend nodes using a fixed scheme — most commonly consistent hashing — computed by clients without a central coordinator.…
• Static stability — Static stability is the reliability principle that a system should continue operating with the last known good state when something fails,…
• Static type specialization — Static type specialization is an interpreter/compiler optimisation that eliminates runtime type dispatching by emitting already-type-specific instructions at compile time.
• Status-based LLM pipeline checkpointing — Status-based LLM pipeline checkpointing is the discipline of tracking each record's processing state in a status column (or equivalent state field) inside the pipeline's tabular…
• Stealth crawler — A stealth crawler is an automated web client that deliberately misrepresents its identity to evade origin-side crawl controls — robots.txt, user-agent allow/deny rules,…
• Step parameter merging — Step parameter merging is the deterministic pipeline that constructs the effective parameter space a workflow step sees at execution time,…
• Step-up authentication — Step-up authentication is an access-control discipline in which a single session carries a gradient of authenticity levels,…
• Sticky partitioner — The sticky partitioner is a Kafka producer-side partitioning strategy that concentrates all unkeyed records from a single client onto one partition at a time,…
• Sticky routing — Sticky routing is the property that the same logical key continues to be routed to the same node across routing-map updates, redeployments,…
• Storage bottleneck migration — Storage bottleneck migration names the recurring observation across the history of high-performance systems: the dominant performance bottleneck in the I/O path is not stable…
• Storage-engine maturity as data risk — Storage-engine maturity as data risk is the framing that the years of production exposure of a database's storage engine is the substrate durability metric for that platform…
• Storage IO latency SLI thresholds — A three-tier ladder of block-device latency thresholds for databases, derived from Zalando's empirical RDS Postgres fleet observations:
• Storage latency hierarchy — The storage latency hierarchy is the gap — measured in round-trip time — between each tier of storage a program can reach.…
• Storage media tiering — Storage media tiering is the architectural practice of deploying multiple distinct storage media types in a coordinated hierarchy,…
• Storage overhead and fragmentation — Storage overhead is the ratio of raw capacity consumed to live data stored. In an immutable / append-only substrate, overhead is driven by two separable forces:
• Storage replication for durability — Storage replication for durability is the architectural pattern where data is written to N independent storage targets — typically a primary + (N−1) replicas…
• Storm-induced infrastructure damage — Storm-induced infrastructure damage is a class of weather-driven failure where severe storms (hurricanes, typhoons, major extratropical cyclones, named European windstorms,…
• Stratified topic sampling — Stratified topic sampling is the production-monitoring sampling strategy that partitions incoming traffic by topic / intent / category and draws a proportional (or…
• Stream adapter overhead — Stream adapter overhead is the allocation / copy / buffering cost of translating between two streaming API conventions — most commonly Node.js streams (stream.Readable) and Web…
• Stream-based file processing — Stream-based file processing is the discipline of processing files one record at a time (or one chunk at a time) rather than loading the full file into memory.…
• Stream replayability for iterative pipelines — Stream replayability is the architectural property that a streaming broker retains historical events long enough and cheaply enough that downstream consumers can re-process…
• Streaming aggregation — Streaming aggregation is the pattern of aggregating metrics in transit — as samples flow from producers to storage — instead of querying raw samples and rolling them up at read…
• Streaming as agile data platform backbone — The structural claim that a modern, AI-native data platform requires a streaming substrate — not a batch-ETL substrate — as its core decoupling layer between producers…
• Streaming model weight update — Streaming model weight update is the index-freshness mechanism used in Meta's SilverTorch retrieval substrate (Source:…
• Streaming SSR — Streaming server-side rendering is an SSR variant where the server emits HTML incrementally as regions of the page become ready, instead of blocking until the full tree renders.…
• Strict eventual consistency — Strict eventual consistency is a refinement of eventual consistency that guarantees convergence to a deterministic state for any quiescent system, given any combination of:
• Strong Consistency (Read-after-Write) — Strong read-after-write consistency is the guarantee that once a write to a key completes, any subsequent read of that key observes the written value…
• Structure-of-arrays decomposition — Structure-of-arrays (SoA) decomposition is the reversible transform that turns an array of records (AoS — each record a struct of fields) into a struct of arrays (SoA…
• Structured journaling tool — A structured journaling tool is an agent tool whose only job is to accumulate typed entries into an append-only log that represents an agent's working memory.…
• Structured output reliability — Structured-output reliability is the quality axis separate from semantic correctness that asks: did the LLM produce a parseable,…
• Stub vs recursive resolver — The stub vs recursive resolver dichotomy describes the two fundamentally different kinds of DNS client software. RFC 1034 §5 uses the word "resolver" for both,…
• Sub-agent — A sub-agent is an LLM agent spawned by another agent, with its own separate context array and typically its own tool allowlist,…
• Sub-topology (Kafka Streams) — A sub-topology in kafka-streams is a connected component of the topology DAG under the graph of shared operators and state stores.…
• Subject Pin — A subject Pin is Pinterest's term for the Pin a user is currently viewing — the immediate visual context driving the page's contextual recommendations.…
• Submarine cable incident — A submarine cable incident is damage to or failure of an undersea fibre-optic cable that carries a non-negligible share of a country's or region's Internet connectivity.…
• Subprocess plugin isolation — Subprocess plugin isolation is the design choice to run a plugin as a separate OS process from the host application, with an IPC protocol (typically gRPC over a Unix domain socket…
• Supertool — A supertool is Zalando's term for "applications or scripts that have the ability to execute large-scale changes across the infrastructure" — including destructive changes.…
• Supervised fine-tuning — Supervised fine-tuning (SFT) is the phase of LLM adaptation in which an already-pretrained base model is trained on labelled (input,…
• Support burden as architecture signal — Support burden as architecture signal is the discipline of treating sustained high customer-experience-ticket volume concentrated on one feature as a diagnostic for structural…
• Surface attribution error — Surface attribution error is a named LLM failure mode where the model "is making decisions based on surface-level clues rather than deeper meaning or causality"…
• Surface-specific calibration — Surface-specific calibration is the use of separate calibration layers for each traffic segment (product surface, view type, user segment, country,…
• Surrogacy in causal inference — Surrogacy is the design stance of estimating the causal effect of an intervention on a long-term outcome by estimating it via a short-term mediator — the surrogate…
• Switch-back experiment — A switch-back experiment is an experimental design that alternates treatment and control over time in the same unit (typically a single market) — e.g.…
• Symbol ID — A symbol ID is a stable, unique string identifier for a named code symbol (class, function, method, type, variable). The load-bearing property is stability under code motion:…
• Symbol renamespacing — Symbol renamespacing is the automated rewrite of a C++ codebase to place all of its symbols inside a flavor-specific namespace,…
• Symbolic-information payload — Symbolic-information payload is the side-table of human-readable names accompanying a profile: function names, file paths, line numbers, executable/binary build IDs,…
• Symptom-based alerting — Symptom-based alerting is the alerting strategy in which the primary alerting rules fire on user-visible symptoms (a business operation's error rate or latency SLO breach,…
• Symptom vs cause metric — A symptom metric summarises the health of a system by measuring the downstream observable of a chain of underlying queues and contention points,…
• Synchronization tax — Synchronization tax is the ongoing engineering + operational cost of maintaining a derived secondary store (most commonly a search index, but also a cache, a read model,…
• Synchronous Block Replication — Synchronous block replication is a storage-layer replication semantic where every write to the primary block device is replicated to a secondary device and acknowledged…
• Synchronous fast path streaming — Synchronous fast path streaming is the streaming-API optimisation technique where an implementation returns an already-resolved Promise (or equivalent) when data is already…
• Synchronous vs asynchronous GPU readback — Readback is copying pixel or buffer data back from the GPU to CPU-accessible memory. Whether readback is synchronous or asynchronous is a fundamental API-design choice that ripples…
• Syscall allowlist — A syscall allowlist is the policy artefact that names the set of system calls a program is permitted to issue after entering a sandboxed mode.…
• System-level playbook scope — System-level playbook scope is the design decision to write incident playbooks at the system altitude (several collaborating microservices) rather than the per-microservice…
• Systems thinking for reliability — Systems thinking for reliability is the engineering-discipline stance that software systems are complex socio-technical systems whose reliability behaviour is non-linear,…
• Table-group ACL — Table-group ACL is a database-authorization model where tables are bucketed into named groups (by exact name or name-prefix pattern) and each group carries three user-list roles:…
• Tag cardinality collapse — The technique of replacing specific tag values with a single placeholder value (typically `) at emission time to bound the number of distinct aggregate records produced…
• Tag protection — Tag protection is a Git-server-side invariant that, once a named tag is published, the tag cannot be deleted or re-pointed to a different commit.…
• Tagged pointer — Pack non-pointer data into the architecturally-unused high or low bits of a machine pointer, then mask/shift at access time to recover pointer and tag separately.…
• Tail-call continuation interpreter — A tail-call continuation interpreter is a bytecode-VM design where each opcode is a free-standing function; each function returns by tail-calling the function for the next opcode…
• Tail category coverage — Tail category coverage is the recsys quality concern that products in sparse / low-volume categories systematically lack representation in retrieval and ranking outputs because…
• Tail latency at scale — "Tail latency at scale" names the failure mode where, as a system fans a single logical operation out across N hosts, the probability that at least one host is experiencing its…
• Tail-latency spike during queueing — A tail-latency spike during queueing is the pathology where a small backend slowdown produces a disproportionately large increase in P99/P99.9 latency because a FIFO queue forms…
• Tainted connection — A tainted connection is a pooled database connection whose per-session state has been mutated by a caller such that returning it unmodified to the pool would leak that state…
• Tape storage (sequential access) — Tape storage is a non-volatile magnetic storage medium where data is recorded on a long strip of tape wound in a cartridge.…
• Task and actor model — The task and actor model is a low-level distributed-compute programming model that exposes two orthogonal primitives:
• TCP three-way handshake — The TCP three-way handshake is the connection-establishment protocol used by TCP before any application data can flow. It guarantees a reliable,…
• TCP tracepoint — TCP tracepoints are stable kernel hook points for observing TCP socket state transitions. They were added to Linux around the 4.15-4.16 era (Brendan Gregg's TCP Tracepoints post…
• Tech-debt compounding — Tech-debt compounding is the self-reinforcing cycle in which version lag on a load-bearing system makes the next upgrade harder than the previous one,…
• Tech Radar as language / framework governance — The Tech Radar model — popularised by ThoughtWorks in 2010 — is a ring-based classification used by engineering organisations to govern the lifecycle of a technology (language,…
• TEE side-channel vulnerability — A TEE side-channel vulnerability is any information-leakage path that bypasses the TEE's intended isolation boundary — typically through microarchitectural state, timing, power,…
• Telemetry as discovery substrate — Telemetry as discovery substrate is the architectural bet that live observability data (metrics + logs + traces) is a sufficient input to discover the shape of a running system…
• Telemetry-based resource discovery — Telemetry-based resource discovery is the technique of inferring runtime relationships between infrastructure resources from observability data (traces, service-mesh traffic,…
• Telemetry buffer-and-flush — Buffer-and-flush is the telemetry pattern in which events are aggregated in-process on each emitter host, and the aggregated state is periodically flushed to the ingestion backend.…
• Telemetry TTL as a one-way door — Telemetry data has a TTL. Metrics, logs, and traces expire on retention windows — vendor-imposed or self-imposed. Once expired,…
• TELeR prompt framework — TELeR is a prompt-engineering taxonomy that decomposes a prompt along four axes — Turn, Expression, Level of details, and Role…
• Temperature-zero for deterministic code generation — Temperature=0 is the LLM-sampling setting that makes the decoder deterministically pick the highest-probability token at every step (greedy decoding),…
• Temporal bucket discretization — Temporal bucket discretization is the technique of segmenting a continuous time range into a contiguous sequence of fixed-size discrete intervals so that independently-produced…
• Temporal locality (recency bias) — Temporal locality (a.k.a. recency bias in the application-caching framing) is the property that recently accessed data is more likely to be accessed again soon than older data.…
• Temporal logic specification — Temporal logic is a family of formal languages for specifying behaviors of systems over time — "X eventually happens", "A always precedes B",…
• Temporal persistence layer — A Temporal cluster is a Temporal Server paired with a persistence layer — the durable data store that holds every piece of state the server needs to survive restarts,…
• Temporal profiling — Temporal profiling is continuous, wall-clock-timestamped CPU sampling retained long enough that, after a rare event occurs,…
• Temporal topology query — Temporal topology query is the ability to ask, of a service dependency graph, "what did this graph look like at time T?"…
• Tenant header routing — Tenant header routing is the architectural pattern where a client's request carries an explicit tenant identifier in an HTTP header,…
• Tenant isolation in routing layer — Tenant isolation in routing layer is the requirement that multiple use cases (tenants) sharing a routing proxy be isolated from each other's failure and load-shape impact.…
• Tenant isolation — Tenant isolation in a multi-tenant SaaS is the property that one tenant's users, data, and policies are strictly inaccessible to other tenants…
• Tenant onboarding time — Tenant onboarding time is the wall-clock duration from "new tenant signed" to "tenant serving production traffic". It is a first-class business and operational metric…
• Tensor parallelism — Tensor parallelism is a multi-GPU model-sharding strategy in which individual weight matrices (tensors) within each transformer layer are split across multiple GPUs…
• Tentative request — A tentative request is one that the leader has transmitted to its followers but has not yet confirmed to meet the durability requirement.…
• Tentative schedule — Tentative schedule is an algorithmic primitive for online scheduling: the scheduler "maintains a tentative schedule by assigning the jobs that have already arrived to disjoint time…
• Terminology Server — A terminology server is the system component that normalises clinical concepts across code systems (LOINC, SNOMED CT, RxNorm, ICD-10,…
• Test Case Minimization — Test case minimization (also: shrinking) is the step in a property-based testing workflow where, once a random input causes a test failure,…
• Test cluster as break-things environment — A break-things test cluster is a pre-production environment that is deliberately scaled, versioned, and configured to match production,…
• Test data generation for edge cases — Test data generation for edge cases is the workflow of creating synthetic dev-environment test fixtures that mimic the long-tail input distributions seen only in production.…
• Test feedback loop — The wall-clock time from push-commit to full-suite pass/fail signal — a DevEx-first-order primitive that directly shapes how engineers develop, rebase, and ship changes.…
• Test parallelism worker count — The number of worker processes a parallelised test runner spreads a test suite across. In Rails + minitest the knob is literal: parallelize(workers: N).…
• Test pyramid — A test pyramid (Mike Cohn, popularised by Martin Fowler) is a shape heuristic for a test suite's composition. From widest (foundation) to narrowest (apex):
• Test reliability through simplification — Test reliability through simplification is the discipline of improving test-suite reliability by reducing the number of scenarios and the number of interactions per scenario…
• Test sensitivity — Test sensitivity (closely related to statistical power) is the ability of an experimentation design to detect a real effect of a given size, at a given significance level,…
• Text-to-SQL — Text-to-SQL is the task of generating an executable SQL query from a natural-language question, given a specific database schema (and, in practice,…
• Text-to-text regression — Text-to-text regression is numeric prediction done by a language model that reads the inputs (x) as a string and emits the target (y) as a decoded string,…
• TGW Appliance Mode — Appliance Mode is a property of an AWS Transit Gateway attachment that pins a flow to a specific Availability Zone's endpoint over the lifetime of the flow,…
• Theta sketch — A Theta sketch is a streaming probabilistic data structure for distinct-value cardinality estimation that additionally supports full set algebra — union, intersection,…
• Third-party caveat — A third-party caveat is a Macaroon caveat that says "this token is only valid if accompanied by a discharge token from authority X." It lets a token issuer delegate a sub-decision…
• Thread (OS) — A thread is a unit of execution within a process that shares the process's address space and code segment with peer threads, but maintains its own program counter, register state,…
• Thread-per-core scheduling — In a thread-per-core runtime (e.g., Seastar), each CPU core owns its data and communicates only via message passing. Shared mutable state is forbidden by design.…
• Thread pool exhaustion — Thread pool exhaustion is the failure mode where every worker thread in an application's request-handling pool is blocked waiting on a slow or unresponsive downstream.…
• MySQL threads_running — threadsrunning is a MySQL server status variable reporting the number of threads currently executing a query. It is distinct from threadsconnected (threads that hold a connection,…
• Threat modeling — Threat modeling is the discipline, originating in security engineering, of enumerating threats against a system before deciding on countermeasures.…
• Three-database problem — The three-database problem is the named infrastructure failure mode for teams building AI agents: they end up running three unrelated storage systems…
• Three-layer agent platform stack — An emerging architectural decomposition for production AI agent platforms into three distinct responsibility layers:
• Three-valued logic — Three-valued logic extends classical boolean logic with a third state — variously called null, unknown, or indeterminate…
• Throttler client identity — Throttler client identity is a (usually hierarchical) identifier that a client presents — or that a throttler reconstructs — for each throttler.check() call,…
• Throttler client starvation — Throttler client starvation is the failure mode where one client's workload pushes the throttling metric continuously above threshold for an extended period,…
• Throttler exemption — Throttler exemption is the design choice to let specific clients — identified by client identity — bypass all or some of the throttler's checks,…
• Throttler fail-open vs fail-closed — When a throttler is unreachable — crashed, partitioned, restarting — clients must decide locally whether to proceed with full power (fail-open) or to consider themselves rejected…
• Throttler hibernation — Throttler hibernation is the design posture of slowing or pausing the throttler's own activity — metric collection, inter-throttler communication,…
• Throttler metric scope — A metric scope is the unit of the system over which a throttling metric is meaningful. Two scopes show up in sharded-MySQL deployments:
• Throttler threshold as standard — When a system runs a workload aggressive enough to push against a throttler's threshold for extended periods, the threshold itself becomes the steady-state metric value.…
• Throughput vs IOPS — Throughput and IOPS are the two independent capacity dimensions of any storage system, and cloud providers provision and price them separately.
• Thundering herd — A thundering herd is a failure mode where a resource is overwhelmed by too many simultaneous requests, typically because many clients were previously blocked / disconnected / idle…
• Tier-based instance sizing — Tier-based instance sizing is the capacity abstraction where a managed service exposes a discrete, ordered catalog of instance / cluster sizes — MongoDB Atlas's M10, M20, M30, …,…
• Tier topology as branch hierarchy — Tier topology as branch hierarchy is the architectural principle that deployment environments (production, staging, UAT, QA,…
• Tiered storage as primary fallback — Tiered storage as primary fallback is the architectural property where a streaming broker / database stores primary (write-path + recent-read-path) data on local NVMe and offloads…
• Tiered storage for fast decommission / recommission — Tiered storage for fast decommission / recommission names the operational benefit of tiered storage that's distinct from its usual cost / capacity framing: because historical…
• Tiered storage (hot / warm / cold) — A storage architecture that splits data across three tiers by age and access pattern:
• Tight migration scope — Tight migration scope is the posture of constraining a large infrastructure migration to the single system being swapped…
• Time-aware LRU cache eviction — Time-aware LRU is LRU plus a per-entry TTL. Entries are evicted either when the cache is full (standard LRU) or when their expiration timer fires, whichever comes first.…
• Time budget sharing — Time budget sharing is the chained-call timeout strategy where a caller's SLA is divided among its sequential downstream calls,…
• Time-series bucketing — Time-series bucketing is the practice of decomposing a continuous stream of time-stamped events or values into discrete, aligned-to-fixed-boundary time windows (buckets)…
• Time to first token (TTFT) — Time to first token (TTFT) is the LLM-serving latency metric measuring the delay between a request arriving at the server and the first output token being emitted back…
• Time to know vs. time to recover — Time to know (TTK) and time to recover (TTR) are two adjacent but distinct incident-response KPIs. They partition the operational lifecycle of an incident:
• Time to react to a relevant quantum event — Time to react to a relevant quantum event is the urgency metric Meta uses to organise the PQC Migration Levels ladder. It measures how quickly an organisation can respond…
• Timing-based information surfacing — Timing-based information surfacing is the UX discipline of showing information at the moment it becomes actionable, not at the moment it first becomes available.…
• TLD-level failure blast radius — TLD-level failure blast radius is the structural property of the DNS hierarchy that makes a failure at a TLD registry affect every domain under that TLD simultaneously,…
• TLS 1.3 0-RTT handshake — TLS 1.3 0-RTT handshake is the TLS 1.3 extension that lets a client resume an established TLS session without a round trip…
• TLS close_notify — closenotify is the TLS protocol's graceful-shutdown signal — an Alert record with description closenotify sent by one peer to inform the other that it will not send any more…
• TMA (top-down microarchitecture analysis) — Top-down microarchitecture analysis (TMA) is a CPU-level performance-diagnostic methodology that uses hardware performance counters to partition every cycle of CPU execution into…
• Token-aware load balancing (LLM serving) — Token-aware load balancing is the admission-control / routing primitive for LLM-serving load balancers in which the balancer's per-endpoint load estimator tracks in-flight tokens…
• Token-bucket slow-query limiter — A token-bucket slow-query limiter is a rate limiter applied to the instrumentation path that records "slow query" events so that bursts are captured but long-tail overall…
• Token-count-based batching — Token-count-based batching is the GPU-inference-serving discipline of grouping pending requests into batches whose total token count is bounded by a fixed budget…
• Token enrichment — Token enrichment is the practice of adding authorization-relevant attributes (roles, group memberships, tenant identifiers,…
• Token-heavy system — A token-heavy system is an LLM-based application whose per-request token consumption is the load-bearing capacity and cost driver — large enough that feasibility, cost,…
• Token-probability as ranking signal — Token-probability as ranking signal is the technique of retaining the LLM's per-token output probability (a logit or a normalised probability) past the discrete decision and using…
• Token vault — A token vault is an out-of-band credential-management component that holds long-lived provider credentials for the enterprise and mints short-lived,…
• Token verification — Token verification is the serving-side primitive that speculative decoding and speculative cascades rely on: given an N-token draft produced by a small drafter model,…
• Tokenize transform — The tokenize transform replaces a stream of repeated values with a pair:
• Tokenized secret — A tokenized secret is a placeholder credential — opaque to the client — that a trusted egress hop substitutes for the real secret at the moment the outbound call leaves…
• Tombstone (deletion marker) — A tombstone is a special marker written in place of a deleted record in an eventually-consistent or append-only store. It says "this key used to exist;…
• Tool overlap poisons agent accuracy — A design lesson for LLM agent tool surfaces: two tools that overlap in scope make the model worse, not better. The model gets confused, calls the wrong one,…
• Tool-selection accuracy — Tool-selection accuracy is an LLM agent's probability of picking the correct tool from its available set for a given sub-task.…
• Tool sprawl — Tool sprawl is the named application-layer failure mode of an enterprise running AI agents at non-trivial scale without a centralised inventory of the tools those agents can call:…
• Top-down vs bottoms-up generation — Top-down vs bottoms-up generation is the architectural choice axis for LLM-based content generation where the output is a structured hierarchy (page with themed sections,…
• Topic name as coordination surface — Topic name as coordination surface is the architectural observation that, in a Kafka-shaped messaging system at multi-repo / multi-team scale,…
• Topic-prefix namespacing convention — Topic-prefix namespacing convention is the operational discipline of encoding the origin cluster into topic and consumer-group names via a short prefix (e.g.…
• Topology-aware blast radius — Topology-aware blast radius is the operationalisation of blast radius as a graph traversal over a service dependency graph rather than a memory-and-spreadsheet exercise.…
• Torn page — A torn page is a disk page (typically 8 KB in Postgres) that was partially written when a crash interrupted the write operation…
• Traceability of retrieval — Traceability of retrieval is the agent- reliability property that the path from a user question to the content the agent reformulated into its answer is human-readable…
• Traffic anomaly — A traffic anomaly is a machine-detected deviation of an observation's current volume or shape from the baseline expected for that observation.…
• Traffic-aware migration throttling — Traffic-aware migration throttling is the operator discipline of scaling down an in-progress schema- migration's resource consumption when production traffic spikes…
• Traffic-aware Pre-Rendering — Traffic-aware Pre-Rendering (TPR) is a deploy-time pre-rendering strategy that uses edge analytics — the CDN's record of which URLs have actually been requested…
• Traffic-cohort segmentation — A blast-radius containment posture where a single critical service is run as multiple independent copies, each handling a distinct cohort of traffic (e.g., free customers vs.…
• Traffic shadowing via ingress — Traffic shadowing via ingress is the practice of duplicating production requests at the ingress layer and sending the copy to a second backend while the original still serves…
• Traffic-source tagging in traces — Traffic-source tagging is the discipline of attaching a traffic-class tag to the root span (and propagating it via baggage / trace context) of every request,…
• Trailing-metric patience — Trailing-metric patience is the organisational discipline of committing to a reliability program whose top-line metric lags project delivery by months,…
• Training checkpoint — A training checkpoint is a periodic dump of the full state required to resume a distributed training job after a failure: model weights, optimizer state (Adam m/v buffers),…
• Training-cutoff dynamism gap — The training-cutoff dynamism gap is the drift between a language model's parametric knowledge — frozen at the training-data cutoff date — and the current state of an external,…
• Training / serving boundary — The training / serving boundary is the organizational and infrastructure split between the fleet that trains a model and the fleet that serves it in production.…
• Training-serving tokenizer skew — Training-serving tokenizer skew is the silent-quality-regression failure mode that arises when the tokenizer used during LLM post-training differs — even subtly…
• Trajectory evaluation — Trajectory evaluation scores an agent on how it investigated, not only whether the final answer was correct. For a given evaluation label:
• Transaction commit delay — Transaction commit delay — also called commit queue delay or commit latency — is the time a transaction spends in the commit/flush queue waiting for its changes to be durably…
• Transactional vector index — A transactional vector index is an ANN index whose mutations obey the hosting database's transactional semantics: inserts / updates / deletes land in the index atomically…
• Transfer learning — Transfer learning is the machine-learning practice of taking a model (or model components) trained on one problem in one data distribution and reusing its learned representations…
• Transformation data model — A transformation data model specifies, for each source system, how each of its columns maps (directly or indirectly) to the logical data model of the golden record ().
• Transient job cluster — A transient job cluster is a compute cluster that is created on-demand for a single job run, used exclusively by that run, and torn down when the run completes.…
• Transitive-dependency override build — A transitive-dependency override build is a build-system configuration that forces a specific version of a library pulled in transitively (via another dependency),…
• Transitive-dependency reachability — Transitive-dependency reachability is the graph-theoretic reason a package's cost depends on how it's imported: importing a package transitively brings in every package it imports…
• Transitive parameter resolution — Transitive parameter resolution is the process of resolving a bound layer property's value by walking a chain of parameter references,…
• Translated Query Parity — Translated query parity is the invariant that a search query translated into another language should carry the same search intent as its source…
• Transparent cluster code distribution — The runtime property where code (module definitions, function bodies, local bindings) written on one node of a cluster becomes callable on every other node of the cluster without…
• Transparent Huge Pages (THP) — Transparent Huge Pages (THP) is a Linux kernel feature (since 2.6.38, merged in 2011) that transparently promotes contiguous 4 KiB pages to 2 MiB pages (x86-64) without requiring…
• Transparent memory offloading — A kernel- or hypervisor-level technique that detects cold memory pages in process address spaces and migrates them to a cheaper,…
• Traveling salesman problem — The traveling salesman problem (TSP) is the classic combinatorial-optimization problem: given a set of cities and the distances between each pair,…
• Treat-as-withdraw — Treat-as-withdraw is the BGP error-handling approach defined in RFC 7606 that revises the original RFC 4271 behavior for malformed UPDATE messages.…
• Tree-structured drafting — Tree-structured drafting is a speculative-decoding drafter-side primitive in which the drafter emits a tree of candidate continuations rather than a single linear sequence of N…
• Tribal knowledge — Tribal knowledge is the set of undocumented domain-specific conventions, invariants, and failure modes that live only in engineers' heads and occasional code comments…
• Trie data structure — A trie (pronounced "try" or "tree") is a tree-based data structure where each node represents a substring prefix and each edge is labeled by a single character.…
• TRIM / DISCARD integration — TRIM / DISCARD integration is the filesystem-to-block-layer signal pathway that lets unused filesystem blocks be communicated to the block device so the block layer can avoid…
• Trimean aggregation — The trimean is a single-number summary of a distribution defined as a weighted average of Q1, the median, and Q3:
• Triple-redundant transmission path — N+2 path redundancy for show-critical live broadcast contribution feeds: every primary member-facing stream leaves the venue over three completely discrete transmission paths,…
• Trunk-based development — Trunk-based development is the branching model in which a single long-lived branch (main / master / trunk) is the source of truth, and every developer merges their small,…
• Trust anchor distribution — Trust anchor distribution is the problem of getting a client to trust a specific public key as the legitimate authority for a particular service, hardware fleet,…
• Trust & Safety classifier — A Trust & Safety (T&S) classifier is a small fine-tuned model sitting in front of a user-facing LLM application that classifies incoming user questions (or prompts) into safety…
• Trusted Execution Environment (TEE) — A Trusted Execution Environment (TEE) is a hardware-enforced isolated execution context whose contents — memory, register state,…
• TSDB scaling bottleneck — Scaling a time-series database is an infrequent, disruptive operation for most companies — you provision for peak, over- provision,…
• TTFB vs TTLB (SSR measurement) — TTFB (time-to-first-byte) and TTLB (time-to-last-byte) are two latency-measurement choices for server-rendered HTTP responses,…
• TTL-based deletion with jitter — TTL-based deletion with jitter is the deletion strategy of marking items as expired now with a per-item randomised TTL, letting background compaction / TTL-GC physically remove…
• TTL-based row deletion — TTL-based row deletion is the retention pattern where a background job (cron, scheduler, pgcron, external loop) runs a DELETE statement with an interval predicate…
• Tunable consistency — Tunable consistency is the property that a database lets applications choose the consistency + durability level per operation, rather than fixing one level for the whole system.…
• Tuple sketch — A Tuple sketch is a streaming probabilistic data structure that combines distinct-value cardinality estimation with per-key metric aggregation in a single mergeable summary.…
• Turbo Module + DI contract — Turbo Module + DI contract is a three-language API contract (TypeScript + Swift + Kotlin) used to let a React Native layer communicate with the surrounding legacy native app while…
• Two-axis autoscaling — Two-axis autoscaling combines horizontal scaling (add/remove replicas) with vertical scaling (adjust per-replica capacity or concurrency limits) in a single coupled control loop.…
• Two-level distributed lock — A two-level distributed lock is a lock hierarchy in which contending workers must acquire a local lock first, then a shared / global lock…
• Two-level map KV model — The two-level map KV model is a key-value data shape of the form
• Two-phase completion protocol — The two-phase completion protocol is the consensus-commit shape in which a leader first transmits a request to followers as tentative and — only after durability is met…
• Two-stage forecast + optimise pipeline — A two-stage forecast + optimise pipeline is an architectural idiom for decision-making systems under uncertainty where:
• Two-tier stream + Iceberg query bridge — A two-tier stream + Iceberg query bridge is the architectural property where a single SQL statement reads transparently across both the live streaming tier (broker log segments)…
• Two-tower architecture — A two-tower (or dual-encoder) model is a retrieval / ranking architecture with two independent neural encoders:
• Type-class-driven random generator — A type-class-driven random generator is the mechanism underlying ergonomic property-based testing: a uniform type-indexed function (Arbitrary, Random,…
• Uber-jar metadata loss — Uber-jar metadata loss is the JVM-specific failure mode where a build process that shades / fat-jars / uber-jars its dependencies…
• UDP middlebox hostility — UDP middlebox hostility names the deployment friction QUIC / HTTP/3 inherit because the internet's network middleboxes — ISPs, enterprise firewalls, load balancers, CDNs,…
• UDP reflection + amplification — UDP reflection+amplification is a volumetric-DDoS technique that exploits three properties of many legacy UDP protocols:
• UI primitives as platform building blocks — A UI primitive in a server-driven or declarative UI system is a leaf-level component that:
• ULID (Universally Unique Lexicographically Sortable Identifier) — A ULID is a 128-bit time-ordered identifier designed (Alizain Feerasta, 2016) as a UUID alternative with three design goals: (1) same 128-bit width as a UUID so it fits the same…
• Uncertainty quantification — Uncertainty quantification (UQ) is the discipline of producing a confidence estimate alongside a prediction — not just what the model thinks but how sure it is.…
• Undeclared crawler — An undeclared crawler is any automated web client not listed in its operator's published crawler directory — no matching user-agent, source IP,…
• Undocumented-behavior drift on version upgrade — Undocumented-behavior drift is a class of version-upgrade regression where callers rely on behavior the old implementation provided but the official spec never promised.…
• Undocumented production change — An undocumented production change is any mutation of production state — code deploy, config push, schema migration, infra tweak, credential rotation,…
• Unhandled Rust panic — A Rust worker thread aborts because a .unwrap(), .expect(), assertion, or explicit panic!() was hit without a handler. Behaviorally equivalent to an unhandled exception in other…
• Unified change stream across shards — A unified change stream across shards is a single consumer-facing ordered stream of change events that papers over the fact that the source database is physically sharded across N…
• Unified connectivity layer — A single private-network routing fabric shared across multiple product surfaces (L4 proxy, L7 proxy, edge compute) so that private connectivity is defined once and consumed…
• Unified context-intent embedding — Unified context-intent embeddings are Pinterest's named contribution to production Text-to-SQL: a single embedding space that indexes natural-language descriptions of the business…
• Unified embeddings — Unified embeddings is a memory-optimisation primitive for recommendation / ranking models where multiple categorical features share a single embedding table,…
• Unified Feature Preprocessing across Training vs Serving — A design requirement in production ML: the preprocessing applied to incoming requests in production must be identical to that applied to the training data…
• Unified Graph (Principled GraphQL) — A unified graph is an organisational discipline that treats the GraphQL schema as one shared surface across the entire company — not one graph per team, per product,…
• Unified interface schema — A unified interface schema is a single machine-readable description of a product's APIs, commands, configuration, and RPC surfaces that is rich enough to serve as the source…
• Unified-library leverage — Unified-library leverage is the strategic payoff of running a monoculture core library across a fleet: a change made once inside that library — instrumentation, a new primitive,…
• Unified metric semantics — Unified metric semantics is the discipline of keeping metric names, label schemas, and metadata dimensions consistent across multiple storage backends so that users don't have…
• Unified model catalog — A unified model catalog is the product-surface property of an AI platform where one catalog, one API, one credential model,…
• Unified parameter protocol — A unified parameter protocol is a platform-side normalisation of the parameter space of a class of models (LLMs, image generators,…
• Unified storage and index — Unified storage and index is the managed-service property that a single write both stores the document and indexes it atomically, with no customer-visible sync pipeline,…
• Uniform buffer — A uniform buffer is a chunk of GPU-visible memory that holds shader input parameters (uniforms — values constant across all invocations of a shader dispatch, like a color,…
• Unit-suffix field naming — Name schema fields with the explicit unit (or concrete semantic qualifier) as a suffix — payloadsizebytes, not payloadsize; timestampmsutc, not timestamp;…
• Universal decoder — A universal decoder is a single decoder binary that decodes every frame a family of compressors has ever produced, regardless of which configuration produced the frame.…
• Universal rendering (Zalando Rendering Engine) — In the Zalando / web-frontend-platform sense, universal rendering (sometimes isomorphic rendering) is the property that one codebase runs in both a Node.js backend (to produce…
• Universal rendering (Google) — Universal rendering is the post-2018 property of Google's indexing pipeline that every HTML page Googlebot successfully crawls is enqueued for full rendering by the Web Rendering…
• Universal resource provisioning — Universal resource provisioning is the abstraction of every external cloud / SaaS resource as a declarative object in one uniform control plane,…
• Unix-socket API proxy — A Unix-socket API proxy is a local IPC endpoint — a Unix domain socket — that a privileged process inside a VM or container exposes to the rest of the guest,…
• Unknown zero enum value — Convention of reserving the 0-th element of every protobuf enum as an explicit UNKNOWN / UNSPECIFIED value so that old consumers reading messages produced by newer schemas (which…
• Unlinkability — Unlinkability is the cryptographic property that a token issuance event and a token redemption event cannot be directly correlated by any party — including the issuer, the origin,…
• Unlogged table (Postgres) — An unlogged table in Postgres is a table declared with CREATE UNLOGGED TABLE … whose writes bypass the write-ahead log (WAL) entirely.…
• Unplanned failover playbook — An unplanned-failover playbook is the operational runbook for promoting a replica to primary when the current primary has crashed, become unreachable,…
• Unsharded-to-sharded migration — An unsharded-to-sharded migration is the specific variant of MoveTables that moves one or more tables from a single-instance (unsharded) keyspace to a multi-shard keyspace within…
• Unsigned right shift (>>>) — Unsigned right shift (Java spelling: >>>) is the bit-shift operation that treats its left operand as an unsigned bit pattern,…
• Unused / reclaimable disk buffer — Unused / reclaimable disk buffer is the deliberate reservation of a slice of a storage device that is never allocated for primary data…
• Unwind safety — Unwind safety in Rust is the discipline that guarantees code's data structures remain in a valid state even if a panic unwinds through them.…
• Up/down migration pair — An up/down migration pair is the authoring discipline, native to most versioned schema migration tools, where every migration file exports two symmetric closures: an up()…
• Uplink saturation from backoff — A failure mode where a retry loop on one dependency accidentally write-amplifies a second dependency, and the second dependency's write path saturates the network uplinks.…
• Upper ontology — An upper ontology — also called a top-level ontology or foundational ontology — is an ontology that defines the most general concepts and relations used across multiple narrower…
• URI template (RFC 6570) — A URI template per RFC 6570 is a URI pattern with {placeholder} expressions that expand into concrete URIs at request time.…
• URL-encoding idiosyncrasy in S3 keys — S3 server access logs write the key field url-encoded. But the encoding is not uniform: for most request types it is double-url-encoded;…
• URL normalization — URL normalization is the transformation of URL variants that resolve to the same underlying content into a single canonical form,…
• Use-directive as compilation marker — Use-directive as compilation marker names the idiom where a string literal placed at the top of a function body (or file) acts as a compile-time flag that tells a compiler plugin…
• USE Method — USE = Utilisation / Saturation / Errors — a methodology for locating performance bottlenecks proposed by Brendan Gregg at Netflix.…
• User Action as Token — User-action-as-token is a recommendation-system framing that treats a user's chronological sequence of actions (bookings, views, searches, clicks, listens, watches,…
• User-agent rotation — User-agent rotation is the practice of switching the User-Agent request header across requests (or in response to origin-side blocks) to evade user-agent-based allow/deny rules.…
• User-controlled log fields — Some fields in access logs / request logs carry unescaped user-supplied bytes: HTTP request URI, Referer, User-Agent, request body fragments.…
• User event sequence — A user event sequence is "an ordered list of recent, relevant events for a user, along with the enrichments (signals) attached to each event" (Source:…
• User-interaction tracing — User-interaction tracing is the creation of tracing spans that begin at the user interaction in the browser (button click, tap, form submission,…
• User Perceived Latency — User Perceived Latency (UPL) — sometimes shortened to perceived latency — is the time from when a user performs an action until they see the resulting content.…
• User-space congestion control — User-space congestion control is the QUIC-enabled architectural shift in which the TCP-era kernel-based congestion-control algorithm is relocated into the application library (QUIC…
• User-split experiment — A user-split experiment is classical A/B testing at the user level: randomise users into treatment and control groups, apply different variants of a feature / policy / prompt…
• utf8mb4 vs utf8 (the MySQL UTF-8 trap) — In MySQL, the character set named utf8 is not actually UTF-8. It is a three-byte-maximum encoding that covers only Unicode's Basic Multilingual Plane (BMP) and silently fails…
• Utilisation-prioritised refresh strategy — Utilisation-prioritised refresh is the operational discipline of recomputing pre-computed reports for multi-tenant SaaS only when a tenant (or scope) crosses a utilisation…
• Utilization vs predictability tradeoff — The utilization vs predictability tradeoff is the long-standing tension in multi-tenant distributed-compute systems: high utilisation comes from packing workloads tightly onto…
• UUID primary-key antipattern — Using a random UUID (v4, v3, v5) as a clustered-index primary key in InnoDB (or any database where the table is organised as a primary-key B+tree) penalises performance on three…
• UUID string representation in MySQL — MySQL has no native UUID column type; UUIDs in MySQL are stored as 36-character strings (VARCHAR(36)) or as 16-byte BINARY(16) with application-side formatting.…
• UUID_TO_BIN swap flag — MySQL 8.0+ ships a built-in helper function UUIDTOBIN(uuidstring, swapflag) that converts a 36-character UUID string into the 16-byte binary representation — and,…
• UUID version taxonomy — A Universally Unique Identifier (UUID) is a 128-bit identifier designed so independent systems can mint unique IDs without coordination.…
• V8 young generation — The young generation (aka "young space", "nursery") is the area of V8's heap where newly allocated objects live initially.…
• Validating admission webhook — A validating admission webhook is a Kubernetes extension point (admissionregistration.k8s.io/v1 → ValidatingWebhookConfiguration) that lets a cluster operator plug custom…
• Valley-free routing — Valley-free routing is the property — formalised in Gao–Rexford (IEEE/ACM Trans. Networking 2001) — that a well-formed BGP AS path,…
• Variable bitrate (VBR) — Variable bitrate (VBR) is a video-encoder rate-control mode in which the encoder's objective is target quality, not target bitrate: the encoder is allowed to raise or lower bits…
• VARIANT type — VARIANT is a column-level data type that stores semi-structured (JSON-shaped) values with their internal structure preserved and queryable,…
• VCS usability — VCS usability is the design axis of a version control system that concerns the user experience of operating it: command surface, default view, mental model, error messages,…
• Vector Embedding — A vector embedding is a dense numerical representation of a piece of unstructured data — text, image, audio, video, or document — produced by an embedding model,…
• Vector Quantization — Vector quantization (in the context of vector search) is compressing embedding vectors from their full-precision representation (typically float32) to a smaller encoding (int8,…
• Vector search has no scale-to-zero — A production-cost observation about hosted vector-search systems: unlike stateless web services or general-purpose serverless compute,…
• Vector Similarity Search — Vector similarity search is the retrieval primitive behind semantic search, recommendation, and RAG: given a query vector and a corpus of vectors (usually embeddings of documents /…
• Vector tiles — A vector tile is a bundle of vector geometry (points, lines, polygons) covering a single cell of a zoomed map grid, serialised as a compact binary blob…
• Vector versioning for deletion — Vector versioning is a technique for making deletion and reassignment append-only in a posting-list-structured ANN index.…
• Vectorized query engine — A vectorized query engine is a query-execution architecture that operates on batches of rows at a time using SIMD / cache-friendly column-major loops,…
• Vendor-neutral evaluation API — A vendor-neutral evaluation API standardises the call surface developers use to invoke a pluggable capability (flag evaluation, tracing, model inference,…
• Vendor string mismatch on fleet config — Vendor string mismatch is the failure mode in which the same logical configuration target — e.g. a specific network boot interface on a specific NIC port…
• Verifiable-test gap (data queries) — The verifiable-test gap is the structural absence of a deterministic oracle for "is this answer correct?" in open-ended data queries…
• Verifiable transparency log — A verifiable transparency log is a third-party-operated append-only ledger that publishes all acceptable artefacts (binary digests, certificates, keys,…
• Verification cache — A verification cache is a local (client-side or edge) cache of token-verification results, letting a verifier skip the round-trip to the token authority for tokens it has already…
• Verified Bots — Verified bots is the general problem of distinguishing legitimate automated clients (search crawlers, AI training crawlers, uptime monitors, archive spiders,…
• Version-aligned compatibility scheme — A version-aligned compatibility scheme numbers a support tool (operator, Helm chart, client library, migration utility, SDK) so its version number matches the underlying system's…
• Version skew in microservice retrieval — Version skew in microservice retrieval is the structural failure mode of multi-service recsys retrieval architectures (Source:…
• Versioned schema migration — A versioned schema migration is a paradigm for managing relational-database schema evolution in which the schema is expressed as an ordered sequence of immutable,…
• Vertical and horizontal autoscaling — Vertical + horizontal autoscaling is the practice of scaling compute capacity along both axes as a single adaptive primitive: add/remove workers (horizontal) and resize the VM…
• Vertical partitioning — Vertical partitioning splits a monolithic database by moving groups of related tables onto separate database instances. Each moved table stays whole…
• Vertical-scaling cost step-function — The cost of vertical scaling a database is a step-function, not a smooth curve: each hardware tier-upgrade is a discrete jump that forces the operator to pay for unused headroom…
• Vertical scaling — Vertical scaling is the discipline of increasing a single server's capacity — more CPU cores, more RAM, faster storage, higher IOPS…
• Vertical sharding — Vertical sharding is the discipline of moving groups of tables — or a single large table — off a monolithic cluster onto a separate cluster dedicated to that group.…
• VGTID — VGTID is Vitess's keyspace-wide progress token in a VStream — a set of per-shard MySQL GTID positions, one entry per shard in the target keyspace,…
• Vibe coding — Vibe coding is the LLM-driven interactive code-generation workflow where a developer (or another agent) converses with the LLM to produce code, the LLM emits code in bursts,…
• Video transcoding — Video transcoding is the act of decoding a source video and re-encoding it to one or more target encodings, optionally changing resolution, codec, framerate, container format,…
• View-based data environment — A view-based data environment populates a new data environment not by copying rows from a source environment but by emitting SQL views that point back at the source environment's…
• View flattening — View flattening is a React Native rendering optimization that omits views from the native view tree when they are deemed unnecessary for the final layout.…
• View-tree traversal — View-tree traversal is walking a hierarchical UI element tree from a root node downward to compute a derived predicate or aggregate.…
• Viewer-friend closeness — Viewer-friend closeness is the ML-estimated strength of relationship between two connected users on a social product, used as a ranking / retrieval signal when deciding which…
• Vindex — A Vindex (short for Vitess Index) is Vitess's mapping from a row's column value to the keyspaceid that identifies the shard the row lives on.…
• VIP address decoupling — VIP address decoupling is the property that clients address a logical use case (e.g. Objective) rather than a concrete cluster VIP (Virtual IP).…
• Virtual file system (for monorepo) — Virtual file system (VFS) for monorepo is a working-copy-layer primitive that presents the full repository shape to the filesystem (so user tools, build systems,…
• Virtual filesystem over SQLite — A design approach where an agent's filesystem is implemented as typed file operations (read, write, edit, search, grep, diff) backed by SQLite storage inside a serverless compute…
• Virtual Kubernetes cluster — A virtual Kubernetes cluster is a Kubernetes-cluster-shaped isolation unit — its own API server, controller manager, namespaces, RBAC, CRDs,…
• Virtual OPA instance per application — Inside one host process (typically an ingress proxy), run many logically isolated OPA instances, one per application, sharing the process's runtime but each with its own policy…
• Virtual sharding — Virtual sharding is the pattern of splitting a large search index into multiple independent clusters (each a full primary-replica Lucene deployment),…
• Virtual-thread pinning — Virtual-thread pinning is the Java 21 failure mode where a virtual thread (VT) cannot unmount from its carrier OS thread while blocked.…
• Virtual thread — A virtual thread (VT) is a JVM-managed lightweight thread that multiplexes onto a small pool of OS-backed worker threads — the carrier threads — via continuations.…
• Virtual waypoint routing — Virtual waypoint routing is the practice of inserting an invisible intermediate stop into a navigation route — not because the route is meaningfully different with the stop,…
• Visibility order vs. commit order — Visibility order vs. commit order names the architectural decision in an MVCC database about whether the sequence in which committed transactions become readable by new snapshots…
• Visual-equivalence mapping — Visual-equivalence mapping is the human-authored correspondence between two UI-component-library attributes that assigns a source value to the target value producing the same…
• Visual quality metric — A visual quality metric gives a numeric score of the perceived visual quality of a video frame or stream — in practice, how much perceptual quality has been lost relative to some…
• Visual-text Relevance Judgment — Visual-text relevance judgment is the LLM-as-judge shape that scores (query, product) relevance using both product data and product images as evaluation context…
• Visually Complete — Visually Complete is an operational predicate marking the moment a user has "seen the content they came for" on a given surface / page.…
• Vitess foreign-key enforcement — Vitess foreign-key enforcement is the layer in Vitess that takes over from InnoDB for any DML that interacts with a foreign-key constraint with a cascading action.…
• Vitess topo-server — The topo-server (topology server) is the shared state store in a Vitess cluster. It is the coupling point between the two layers of the Vitess proxy tier:
• VLM as image judge — VLM-as-image-judge is the evaluation pattern where a vision-language model (VLM) scores a generated image against a curated list of natural-language evaluation questions,…
• VM deoptimization — Deoptimization is the mechanism by which a type-specialized interpreter or JIT bails out of optimised code when a runtime condition invalidates a compile-time assumption,…
• VM Escape — A VM escape is when code running inside a guest virtual machine breaks through the hypervisor boundary and obtains execution or data access on the host system (or,…
• VM lifetime prediction — VM lifetime prediction is the problem class of predicting how long a virtual machine will run — from creation to shutdown — as an input to placement scoring in a cloud scheduler.…
• Vocabulary bottleneck — The vocabulary bottleneck is a structural failure mode of recsys retrieval architectures that score / score-to-distribute over a fixed atomic-item-ID vocabulary.…
• Vocabulary padding for CUDA kernel selection — Vocabulary padding for CUDA kernel selection is the technique of rounding an LLM's vocabulary size up to the nearest multiple of a hardware-friendly boundary (Netflix uses 64) so…
• Volatile-only prompt-cache isolation — Volatile-only prompt-cache isolation is the multi-tenant security shape for a prompt-caching layer composed of three orthogonal properties:
• VPAT (Voluntary Product Accessibility Template) — A Voluntary Product Accessibility Template (VPAT) is a standard-format document that describes how a product conforms to accessibility standards (typically WCAG 2.x A/AA…
• VPC resolver packet rate limit — The AWS VPC resolver (a.k.a. AmazonProvidedDNS, reachable at the VPC base IP + 2 — e.g. 10.0.0.2 if the VPC base is 10.0.0.0) is rate-limited to 1,024 packets per second per…
• VPN-to-ZTA migration — VPN-to-ZTA migration is the motion of transitioning a corporate network from a perimeter VPN (broad network access behind a single authenticator) toward Zero Trust Architecture…
• VTGate query planner — The VTGate query planner is the component inside Vitess's proxy tier (vtgate) that takes an incoming MySQL-protocol SQL statement and produces an execution plan…
• WAF attack score — WAF Attack Score is Cloudflare's ML-based request classification layer that assigns a score (1–99) to every HTTP request based on structural similarity to historical attack traffic…
• WAL record granularity — WAL record granularity is the property that a database's Point-in-Time Recovery target-time cannot be finer than the cadence at which its Write-Ahead Log flushes durable records.
• WAL replication — WAL replication is the technique of shipping a primary's write- ahead log (WAL) to one or more replicas and having the replicas apply it to produce an identical logical state.…
• Write-Ahead Logging (WAL) — Write-Ahead Logging (WAL) is the durability primitive under nearly every production-grade relational database, distributed log, and many object stores.…
• Warm isolate routing — Warm isolate routing is the scheduling policy used by V8- isolate-based serverless runtimes (canonically Cloudflare Workers,…
• Warm node pool — A warm node pool is a set of pre-provisioned Kubernetes nodes with the base runtime image already pulled, maintained by a predictive algorithm ahead of demand.…
• Warm pool instances — Warm pool instances are compute instances that a serverless substrate keeps alive between jobs instead of deprovisioning them immediately after a job finishes.…
• Warm Sprite pool — A pool of pre-created, pre-paid-cost VMs maintained on every physical worker in the platform fleet, holding VMs in a ready-to-assign state such that a user's create request…
• Warn mode vs enforce mode — Warn mode vs enforce mode is the two-state operational lifecycle for any threshold-based control (resource budget, rate limit, throttler, rule engine,…
• Wasm Git server — Wasm Git server names the implementation shape in which the server side of the Git smart-HTTP protocol runs as a WebAssembly module inside a serverless / edge runtime,…
• Watchdog REPL-channel liveness probe — A REPL-channel liveness probe is the technique of instrumenting a long-running process with an in-process control REPL (read-eval-print loop accessible via local socket,…
• WCAG 2.1 A / AA scope — WCAG 2.1 Levels A and AA is the conventional scope-picker for automated accessibility checks at organisations that take a11y seriously but haven't opted into AAA compliance.…
• WCAG platform applicability gap — WCAG platform applicability gap is the observation that the Web Content Accessibility Guidelines (WCAG) are — as the name says — web-centric,…
• Weakly-adversarial critic — A weakly-adversarial critic is an agent positioned to audit peer agents' work for hallucinations, analytical gaps, and interpretation variability — pointed at them,…
• WebAssembly — WebAssembly (often WASM) is a W3C-standardized binary instruction format for a stack-based virtual machine, designed as a portable compilation target for high-level languages (C,…
• Web-search telephone game — The web-search telephone game is the failure mode in LLM pipelines that use web-search RAG where a smaller summariser model sits between the raw web-search results and the parent…
• Web Streams as SSR bottleneck — Web Streams as SSR bottleneck names the empirical finding — disclosed independently by Cloudflare and Vercel during 2025-2026 benchmark cycles…
• WebAssembly Exception Handling — WebAssembly Exception Handling is a Wasm specification proposal adding structured exception primitives — try, catch, catchall, throw, rethrow — to the Wasm instruction set.…
• Webhook URL validation — Webhook URL validation is the set of submission-time checks a webhook-sender service runs against a user-supplied URL before storing or dispatching to it.…
• Weekly integrity reconciliation — Weekly integrity reconciliation is the pattern of running a lower-frequency, out-of-band audit that checks whether the invariants a fast-path automation is supposed to maintain…
• Weighted Boxes Fusion (WBF) — Weighted Boxes Fusion (WBF) is a bounding-box-detection post-processing technique that merges overlapping bounding boxes…
• AWS Well-Architected Framework — The AWS Well-Architected Framework is AWS's design-review methodology: a six-pillar checklist used to evaluate and improve the architecture of cloud workloads.…
• Well-known URI (RFC 8615) — A well-known URI is a URL with a path starting /.well-known/ reserved for site-wide metadata resources defined by IETF or industry-consensus standards.…
• Well-specified target API — An API is well-specified (from an AI-coding perspective) when:
• What Not To Flag Prompt — "What NOT to flag" is the prompt-engineering discipline of spending more explicit instruction on what an LLM should skip than on what it should do.…
• Whitelist Internet access — Whitelist Internet access is a censorship architecture in which connectivity is granted by explicit allowance, not denial: only a curated list of approved domains / services…
• Whole-article retrieval — Whole-article retrieval is the RAG design choice where the retrieval unit is the entire article, not a paragraph chunk or a sentence.…
• Whole-AZ network partition simulation — Whole-AZ network partition simulation is the chaos-engineering drill that programmatically disconnects an availability zone's network from the rest of the cluster…
• Wide-and-Deep architecture — Wide-and-Deep is a hybrid neural-network architecture for recommendation / ranking tasks (notably CTR prediction) that combines two parallel paths over the same input:
• Wide partition problem — A wide partition is a Cassandra (or more generally, any partition-key-oriented) storage anti-pattern where a single partition grows so large that reads against it become slow,…
• Wildcard-prefix non-sargable predicate — A wildcard-prefix LIKE predicate — a LIKE pattern that begins with % or ` (e.g. '%ron', '%oh%') — cannot use a B+tree index,…
• Window virtualization — Window virtualization (a.k.a. virtual scrolling / virtual lists) is a front-end rendering technique that keeps only the visible portion of a long list or grid in the DOM at any…
• Winning-indicator t-test — In interleaving testing of two rankings A and B, each search (or user) produces a winning indicator — a per-unit signed value:
• WiredTiger cache — The WiredTiger cache is the in-memory buffer pool that MongoDB's WiredTiger storage engine uses to hold uncompressed data + index pages.…
• WireGuard handshake — The WireGuard handshake is the exchange of UDP packets that establishes a WireGuard session. Specified in detail in the WireGuard paper. Properties relevant to systems design:
• WireGuard mesh topology — WireGuard mesh topology is the architectural pattern of connecting WireGuard peers to each other rather than through a central hub/concentrator.…
• Word Error Rate (WER) — Word Error Rate (WER) is the canonical metric for automatic speech recognition quality:
• Work item as agent prompt — Work-item-as-agent-prompt is the architectural framing in which a ticket / work item in an issue tracker (Jira, Linear, GitHub Issues) is treated as a structured prompt for an AI…
• work_mem multiplication — workmem multiplication is the geometric amplification that turns a modest per-allocation cap (e.g. 4 MB) into potentially gigabytes of transient RSS under load.…
• Workflow aggregated view — A workflow aggregated view reconstructs the terminal status of every step across multiple runs of a single workflow instance…
• Workflow breakpoint — A workflow breakpoint is an orchestrator-level pause-at-step primitive — the workflow equivalent of a debugger breakpoint.…
• Workflow compensation action — A workflow compensation action is a paired undo method attached to an action that the workflow engine can invoke in reverse order to walk a partially-completed workflow back…
• Workflow determinism requirement — The workflow determinism requirement is the invariant that a workflow method, given the same inputs, checkpointed action results, and state fields,…
• Workflow orchestration — Workflow orchestration is the infrastructure concern of scheduling, executing, monitoring, and managing multi-step computational workflows…
• Workflow replay from checkpointed actions — Workflow replay from checkpointed actions is a durability mechanism where a workflow engine recovers from a crash by re-executing the workflow method from the beginning,…
• Workflow rollup — A workflow rollup is a recursive summary of step statuses across a workflow instance, flattening nested subworkflows + foreach iterations into a count of leaf steps by status.…
• Workflow run strategy — A workflow run strategy is the policy that determines whether and when a newly-triggered workflow instance actually runs,…
• Workflow signal — A workflow signal is an externally-sent message that mutates state fields inside a running or hibernating workflow, unblocking waitUntil conditions and resuming workflow execution.…
• Working-set memory — Working-set memory — the subset of a database's data + indexes that is actually accessed by the live query workload within a given time window and therefore needs to reside…
• Workload-aware routing — Workload-aware routing is the architectural pattern of making load-balancer / gateway routing decisions based on the shape of the incoming request (query content, target tables,…
• Workload identity — A workload identity is a stable, fine-grained identifier naming the logical unit of software running on a host — typically more specific than "this instance" (AMI/VM)…
• Workspace-scoped to org-wide migration — Workspace-scoped to org-wide migration is the specific multi-tenant re-architecture in which an enterprise SaaS product — originally designed around the assumption that a user…
• Write amplification — Write amplification (WA) is the ratio of physical bytes written to storage to logical bytes the application intended to write.
• Write churn — Write churn is the rate at which a table's rows are created, modified, and removed, measured independently of the table's steady-state size.…
• Write-dependency graph — A write-dependency graph is a server-side (or client-side) data structure that tracks, for every node in an editable document,…
• Write endurance (NAND) — Write endurance of NAND flash is the number of Program/Erase (P/E) cycles a cell can survive before its bit-error-rate (BER) crosses the ECC-correctable threshold.…
• Write-path replication — Write-path replication is the ingest-side pattern where each write is replicated N-way across ingester/writer nodes before the write is considered durable and acknowledged…
• Write-through cache — A write-through cache is a caching policy in which every write goes through the cache and the authoritative backing store in order,…
• Write throughput ceiling — Write throughput ceiling is the point at which a database primary can no longer absorb incoming writes at the rate the workload demands.…
• X-Opaque-Id client attribution — X-Opaque-Id client attribution is the discipline of propagating a unique, caller-supplied identifier on the HTTP X-Opaque-Id request header so that server-side observability…
• xDS protocol — xDS is Envoy's family of dynamic-configuration APIs: a streaming gRPC protocol over which a control plane pushes cluster, endpoint, listener, route,…
• xmin / xmax row versioning (Postgres) — In Postgres, every row version (tuple) on disk carries two hidden metadata columns used to determine which transactions may see it: xmin and xmax.…
• XML signature wrapping — XML Signature Wrapping (XSW) is a family of attacks against XML-DSig verification in which the signed bytes and the interpreted bytes diverge.…
• You build it, you run it — "You build it, you run it" is the ownership model where the team that develops a service is also responsible for operating it — including on-call, incident response,…
• Z-Ordering — Z-Ordering is a multi-dimensional clustering technique on Delta Lake that uses a space-filling curve (the Z-order curve) to map multiple-column key tuples onto a linear ordering,…
• Zebra-not-horse heuristic — The zebra-not-horse heuristic is the oncall-investigator's bias-checker for ruled-out-all-the-common-causes situations. It inverts the clinical-diagnostics aphorism
• Zero-code-change migration — Zero-code-change migration is the organisational and engineering commitment that, when a platform team moves user workloads from one execution substrate to another,…
• Zero-copy data sharing protocol — A zero-copy data sharing protocol lets a recipient (another organisation, team, or region) read live tables in the provider's storage without copying the physical data first.…
• Zero-copy parsing — Zero-copy parsing is the technique of extracting structured data from a serialized byte stream without allocating new memory buffers or copying bytes out of the original…
• Zero-copy sharing — Zero-copy sharing means two or more processes (or languages running inside one process) read the same in-memory data without serialization, deserialization,…
• Zero-ETL operational↔analytical — Zero-ETL operational↔analytical is the architectural property that operational data (in a transactional store) is available for analytical workloads (in an analytical store /…
• Zero-knowledge proof — A zero-knowledge proof (ZKP) is a cryptographic protocol by which a prover convinces a verifier that a statement is true without revealing any information beyond the fact that it…
• Zero-trust authorization — Zero-trust authorization is the design principle that every tier that handles a privileged request must independently verify authorization…
• Zombie exclusion-list state — Zombie exclusion-list state is the class of correctness bug in which a multi-step state mutation includes a mark-excluded-then-act-then-clean-up pattern,…
• Zombie memory cgroup — A zombie memory cgroup ("zombie memcg") is a Linux memory cgroup that userspace has destroyed — the container or process that owned it is gone,…
• Zone-aware shard-allocation stuck drain — The zone-aware shard-allocation stuck drain is the failure mode where an Elasticsearch node drain makes no forward progress because the node being drained is the only copy…
• Zone.js monkey-patching — Zone.js (npm: zone.js) is a library originally built for Angular that provides asynchronous execution contexts ("zones") in JavaScript by monkey-patching all global async…
• Zone-redundant storage — Zone-redundant storage is durable storage whose data is automatically replicated across multiple availability zones at the storage-service layer…