Eve library

svc-study-workspace deployment

Other

Revision 1

On this page

ECS Fargate service following the shared deploy-ecs.yml conventions (family <env>-svc-study-workspace, log group /ecs/oshun-<env>/svc-study-workspace, image <ecr>/oshun/svc-study-workspace:<tag>, built with docker/Dockerfile.node).

The build works, as of 2026-08-15, and did not before. That is worth stating because none of the prerequisites below mentioned it: nx build failed with TS6059 on every experience-plane module the service imports — the @nx/js:tsc executor pins rootDir to the app and each of those libraries is a SOURCE library — so there was no artefact for any of this to deploy. It builds with tsup now, into the package-local dist/ shape Dockerfile.node prefers, with dist/index.js as the entrypoint its CMD already looks for. Every npm package the bundle reaches is declared here and held by src/packaging/artifact-dependency-closure.spec.ts, because bundling turns a library's dependencies into this service's and pnpm will not resolve one that was never declared.

Registration is deliberately not wired yet. Enabling deployment requires, in order:

  1. The YSD-0120/0122 blocking decisions approved (node tools/yemaya-study/check-decision-log.mjs --enforce-rollout must pass — this command is the rollout gate and MUST be a job step in any workflow that deploys this service).
  2. Adding the service to the deploy-ecs.yml resolve case statement: svc-study-workspace) NAME="@yemaya/svc-study-workspace"; PATH_="apps/yemaya/svc-study-workspace" ;;
  3. Terraform service/target-group entries via infra/terraform/ecs-services (YSD-1010 lands the CI/CD wiring).

Container contract#

  • Listens on PORT (default 4020), host 0.0.0.0.
  • Liveness: GET /health → 200 (process identity, version, uptime).
  • Readiness: GET /ready → 200 only when every configured dependency probe passes; 503 otherwise. ALB health checks should target /ready.
  • Metrics: GET /metrics (Prometheus text format).
  • Shutdown: SIGTERM → drains in-flight requests for SHUTDOWN_GRACE_MS (default 10s, hard-capped at 25s to stay inside the ECS 30s stopTimeout), closes the Postgres pool, exits 0.
  • Configuration: validated at startup (src/config.ts); the task fails fast with a readable report on missing/inconsistent values. Production requires DATABASE_URL (from Secrets Manager via the task definition) and structured JSON logs.