Lab 15 — Production-style worker service

Goal

Bounded pool, visibility heartbeat, graceful shutdown, metrics — the whole thing.

Setup

To be written — CLI commands and an IaC equivalent for every resource, all prefixed sqs-course-.

Steps

1. Build the consumer from Lab 09 into a bounded worker pool
2. Add a heartbeat that extends visibility for long-running messages
3. Add a SIGTERM handler that stops polling, drains and batch-deletes
4. Deploy to ECS Fargate with stopTimeout longer than p99 processing
5. Deploy a new revision without the SIGTERM handler; count the redeliveries
6. Redeploy with it; count again
7. Add a liveness probe that fails when the poll loop stalls

What you should observe

• A deploy without graceful shutdown produces one redelivery per in-flight message
• The grace period must exceed processing time to be worth anything
• A naive health check cannot see a wedged poll loop

Why it happened

To be written — ties each observation back to the mechanism in 22 — Self-Managed Consumers on ECS, EKS and EC2.

Experiments to try

To be written — deliberate variations that change the outcome.

Teardown

To be written — every resource created above, in dependency order.

← Back to: 29 — Hands-on Labs