SYSTEM Cited by 4 sources

Amazon ECS (Elastic Container Service)

Amazon Elastic Container Service (Amazon ECS) is AWS's proprietary container-orchestration service, the substrate for both App Mesh (deprecated) and ECS Service Connect (current). Compared to Kubernetes (which AWS also offers as EKS), ECS is AWS-native with tighter IAM / VPC / ALB integration and a simpler abstraction model.

Stub page — minimal viable for the App Mesh discontinuation ingest. Expand on future ECS-internals sources.

Core abstractions

• Task Definition — template for a container group (image, resources, env, IAM role, networking mode).
• Task — running instance of a Task Definition; the unit of compute.
• Service — long-running managed group of identical Tasks with desired-count, health, and replacement policy.
• Cluster — logical group of Services / Tasks.

Tasks can run on EC2 capacity (customer-managed instances) or on Fargate (serverless, AWS-managed compute).

Role in service-mesh story

The ECS Service is the atomic unit for mesh membership:

• In App Mesh, each Service's Task Definition includes a self-managed Envoy sidecar container.
• In Service Connect, each Service's Task Definition cannot also be in App Mesh — the mesh membership is exclusive, which is why migration is forced to blue/green recreate.

Related

• systems/aws-app-mesh — deprecated sidecar-mesh layer above ECS
• systems/aws-ecs-service-connect — current managed mesh layer

Seen in

• sources/2026-05-12-aws-building-hybrid-multi-tenant-architecture-for-stateful-services — canonical wiki instance of ECS cluster as the tenant-isolation boundary for a stateful multi-tenant service. AWS's ad-serving platform runs one dedicated ECS cluster per tenant inside shared AWS accounts; each cluster loads only its tenant's in-memory state, eliminating cross-tenant heap sharing. Canonical dedicated-ECS-cluster- per-tenant pattern with naming convention (tier-1-cell-1-ig-1-tenant-a), task-definition TENANT_ID env var propagation, and ECS-task-per-service 5,000 ceiling applying per tenant because cluster is single-tenant. Capacity math (up to 5 clusters per tenant, up to 100 ECS clusters per infra group) and the cluster- level tenant isolation framing first canonicalised on this ingest.
• sources/2025-01-18-aws-app-mesh-discontinuation-service-connect-migration — the substrate under both meshes; the ECS Service's exclusive mesh-membership constraint is load-bearing for migration.
• sources/2024-08-08-figma-migrated-onto-k8s-in-less-than-12-months — ECS as the origin substrate in Figma's 12-month migration to EKS. Figma enumerates ECS limitations that drove the move: no StatefulSets (had to write custom etcd cluster-membership code on ECS, "fragile and hard to maintain"); no Helm support (OSS like systems/temporal required hand-porting into systems/terraform); poor graceful-node-drain on ECS-on-EC2 vs EKS cordon-and-drain; limited auto-scaling vs CNCF systems/keda + systems/karpenter; missing service-mesh off-the-shelf options; expected slower investment vs vendor-agnostic Kubernetes.
• sources/2026-04-08-aws-build-a-multi-tenant-configuration-system-with-tagged-storage-patterns — ECS tasks on Fargate in private subnets are the substrate for the NestJS gRPC Order Service + Config Service in the multi-tenant tagged-storage architecture; ECS Service as the registration unit discovered by Cloud Map for the event-driven refresh Lambda's gRPC fan-out.
• — ECS as the non-Kubernetes substrate that Zalando Payments' Load Test Conductor scales in lockstep with Kubernetes during an end-to-end load-test run. One declarative load-test API call scales applications across two Kubernetes node pools and an ECS cluster simultaneously — an instance of multi-substrate parallel orchestration from a single control plane for a heterogeneous microservice landscape.