fix(docker): pin standalone bind host against runtime HOSTNAME injection

[jee7s]

Problem

v1.0.0 prod deploy failed at Deploy to ECS (run 29422682506): every task exited 1 at boot with

⨯ Failed to start server
Error: getaddrinfo EAI_AGAIN default

Root cause

The Next standalone server.js binds to process.env.HOSTNAME || '0.0.0.0'. The Dockerfile sets ENV HOSTNAME=0.0.0.0, but ECS Managed Instances host-network tasks set a container hostname at create time, which injects HOSTNAME=default into the runtime env over the image ENV. The server then tries to bind to a host literally named default → DNS failure → exit 1 → crash loop → services-stable waiter timeout.

Dev never hit this: Fargate/awsvpc doesn't support container hostnames, so nothing overrides the image ENV.

Fix

Pin the bind address in the CMD itself — sh -c "HOSTNAME=0.0.0.0 exec node …" — so no runtime injection can override it. exec keeps node as PID 1 for signal handling. The migrate task overrides the command entirely and is unaffected.

Verified: env HOSTNAME=default sh -c 'HOSTNAME=0.0.0.0 exec node -p "process.env.HOSTNAME"' → 0.0.0.0; CI's deploy-build smoke test boots the image with the new CMD.

After merge: delete + re-cut the v1.0.0 release (third time — migrations already applied and are idempotent).

🤖 Generated with Claude Code

[jee7s]

Superseded by infra#124 — same root cause, fixed via task-def environment instead so dev and prod env definitions stay identical (per review discussion; task-def env wins over the runtime injection, vercel/next.js#58657).