All new deployments fail healthcheck with "service unavailable" even though the application responds HTTP 200 to /health internally.

Symptoms:

• App starts, listens on 0.0.0.0:3000, responds 200 to /health (visible in deploy logs)

• Railway's healthcheck prober simultaneously reports "service unavailable" for every attempt

• Retries exhaust the full window and deploy fails

Deploy logs (truncated):

Starting Healthcheck
Path: /health
Retry window: 1m40s

Attempt #1 failed with service unavailable. Continuing to retry for 1m29s
Attempt #2 failed with service unavailable. Continuing to retry for 1m27s
timestamp=... level=INFO message="Listening on http://0.0.0.0:3000"
timestamp=... message="Sent HTTP response" http.status=200 http.method=GET http.url=/health
timestamp=... message="Sent HTTP response" http.status=200 http.method=GET http.url=/health
Attempt #3 failed with service unavailable. Continuing to retry for 1m25s
Attempt #4 failed with service unavailable. Continuing to retry for 1m21s
...
Deploy failed

What I've tried (all produce the same result):

• railway up (CLI upload)

• railway redeploy (build from GitHub repo)

• Removing Docker HEALTHCHECK instruction from Dockerfile

• Removing PORT env var (let Railway inject it)

• Setting PORT=3000 explicitly

• Increasing healthcheckTimeout to 300s in railway.toml

• Disabling Cloudflare proxy (DNS only)

• Removing the auto-generated Railway domain

• Deleting the service and creating a brand new one

What works:

• Redeploying a previously successful deployment from the dashboard (old cached image)

Setup:

• Node.js 24.13.1-alpine, Effect Platform, Prisma ORM

• PORT=3000, app listens on 0.0.0.0:3000

• Custom domain with target port 3000

• railway.toml: healthcheckPath = "/health", healthcheckTimeout = 300

This looks identical to the issue in this thread which was caused by conflicting firewall rules and required a platform-side fix.