Two services in one project, same repo (apps/api), same Docker image,

same start command. One works, one never becomes healthy.

Start command:

npx prisma migrate deploy && echo '>>> STARTING NODE' && node dist/main.js; echo ">>> EXITED code=$?"

The Prisma output appears in the deploy logs. Neither echo ever does —

including the second one, which runs unconditionally via ";". The command

is visibly applied in Settings > Deploy > Custom Start Command.

The failing service then produces zero application output. It's a NestJS

app that normally logs ~500 lines at boot. Healthcheck retries the full

window and the deploy fails with "1/1 replicas never became healthy".

Deploy log, in full:

Starting Container

Prisma schema loaded from prisma/schema.prisma

Datasource "db": PostgreSQL database "railway" ...

31 migrations found in prisma/migrations

No pending migrations to apply.

npm notice ...

That's the entire log. Nothing after.

The sibling service — same image, same start command, but no healthcheck

and no public domain — boots fully and logs everything normally.

The app also starts fine locally using the failing service's exact

variable set, and answers /api/v1/health with 200.

Already tried:

• Detached the volume (was mounted at /data) — no change
• Changed UPLOAD_DIR to a container-owned path — no change
• Set PORT explicitly to 3000 (it was not in Railway's injected vars) — no change
• Verified all env vars are present
• Build succeeds; image builds clean

Questions:

1. Why would echo produce no output when prisma in the same command does?

2. Is stdout lost when a container is terminated for a failed healthcheck?

3. healthcheckPath is set in railway.toml but Settings > Healthcheck Path

   displays as unset — is config-as-code applying it?