STAGING-ONLY issue. Production has not been modified and remains healthy.

ENVIRONMENT

Project: talented-commitment

Project ID: f3a62cca-b1cd-47a6-a584-d4cff78aaccc

Environment: staging

Environment ID: a3798345-d7ca-4cd8-b47c-c0b5fe5f6fe0

Service: web

Service ID: 8a124d90-d640-4ce7-b0a7-6af938d437ea

Plan: Hobby

ISSUE

Our pre-deploy command consistently fails after a successful build with only:

"Pre-deploy command failed"

No stdout, stderr, exit code, or actionable error is shown.

GraphQL deploymentEvents reports:

step: PRE_DEPLOY_COMMAND

payload.error: "Pre-deploy command failed"

deploymentLogs/buildLogs contain no pre-deploy output through CLI or GraphQL.

PRIMARY COMMAND

cd /app/packages/db && node dist/scripts/migrate.js

TESTS

1. Normal migration command

Deployment: ceef2110-5d64-4ac4-9181-996bd9b85197

Default Railway timeout.

Failed after ~7.9s.

1. Trivial node -e diagnostic checking only DATABASE_URL hostname

Deployment: d3ce46d5-d4c0-4006-9825-68b651f62c86

preDeployTimeoutSeconds=60.

Failed after ~6.0s.

1. Explicit shell wrapper:

/bin/sh -c "cd /app/packages/db && node dist/scripts/migrate.js"

Deployment: 5e9972c5-e14d-4d89-bd54-c445f13bc0ec

preDeployTimeoutSeconds=60.

Failed after ~10.2s.

VERIFIED

• Exact migration command succeeds immediately through railway ssh.
• Migration files/dependencies are present in the runtime image.
• DATABASE_URL is available and DB is reachable from the running container.
• 60s timeout did not change the failure.
• Explicit /bin/sh -c did not help.
• Trivial diagnostic also failed in PRE_DEPLOY_COMMAND.
• Application builds succeed.
• Existing staging deployment remains healthy; /api/health/ready returns 200.
• Migration ledger remained 16/16 through all tests.
• 0015 remains applied exactly once.
• No migrations were pending and no partial DB changes occurred.
• Production was never modified.
• Staging is restored to the intended fail-closed migration command.

EXPECTED

Build succeeds -> pre-deploy migration command executes -> no pending migrations -> command exits successfully -> deployment continues.

OBSERVED

Build succeeds -> PRE_DEPLOY_COMMAND -> fails after ~6-10s with only "Pre-deploy command failed." New deployment never becomes active.

QUESTIONS

Has anyone seen this behavior with Railway pre-deploy commands, particularly Dockerfile deployments or Hobby plans?

We're trying to identify anything we've missed before concluding this may be Railway-side.

1. Is there anything incorrect about our pre-deploy command?

2. Are there differences between railway ssh and the pre-deploy environment that could explain why the command works via SSH but fails here?

3. Does DATABASE_URL/private-network access require anything additional during pre-deploy?

4. Are pre-deploy commands fully supported on Hobby?

5. Is there a way to retrieve stdout/stderr or the real process exit code when Railway only reports "Pre-deploy command failed"?

6. Is there another safe diagnostic that would distinguish a configuration issue from a Railway execution-environment issue?

I've stopped further staging experiments pending guidance. If I'm missing a configuration requirement, we'd appreciate help identifying it. If this looks Railway-side, guidance from Railway staff on the next step would be appreciated.