Skip to content

PATTERN Cited by 1 source

Adaptive rate control

A pattern for background work (migrations, compaction, replication) that must coexist with latency-sensitive foreground traffic on the same cluster. The controller dynamically adjusts throughput based on real-time signals (rejections, backpressure, latency) rather than using fixed rate limits.

Variants in AOSC

Controller Strategy Use case
Fixed Static batch size + concurrency Offline migrations where max speed is desired
Adaptive batch (default) AIMD on batch size; reacts to backpressure/admission-control rejections Safe default for live clusters
Adaptive Gradient2 tuning of both batch size and concurrency Aggressive but cluster-aware; maximize throughput within headroom

Design Principles

  • Start slow, converge to capacity. Additive increase ensures gradual ramp-up; multiplicative decrease (AIMD) ensures fast back-off on overload.
  • React to system signals, not wall-clock estimates. Admission-control rejections and HTTP 429s/503s are direct evidence of overload.
  • Separate control per phase. Backfill (bulk, sequential) and replay (streaming, incremental) may need different rate profiles.

Seen In

(Source: sources/2026-07-24-atlassian-online-index-migration-and-shard-scaling-in-opensearch-with-aosc)

Last updated · 596 distilled / 1,814 read