Hello Railway team,

I am seeing repeated Worker image-build failures that occur before any Dockerfile instruction begins.

Workspace plan:

Hobby

Project:

khidyao-content-studio-staging

Environment:

staging

Service:

@khidyao/worker

Service ID:

1518ebdc-fb0e-426c-8a8b-8ee477104ce4

Region:

asia-southeast1-eqsg3a

Target commit:

6ec87a0e53810911ff0198800bdc5433e7d01cac

Failed deployment IDs and timestamps:

1. eccce0dc-ab47-45a0-a9e4-aa34652612b3

   Created: 2026-08-25T23:58:27.260Z

2. 2535a617-f805-4077-a45a-41483742f04c

   Created: 2026-08-26T04:12:21.930Z

   BUILD_IMAGE: 2026-08-26T04:12:23.691Z

            through 2026-08-26T04:12:29.288Z

   Terminal failure: 2026-08-26T04:12:29.305Z

Assigned Metal builder:

builder-nxrjjq

Observed behavior:

• SNAPSHOT_CODE completes successfully.

• BUILD_IMAGE begins and fails approximately five to six seconds later.

• The complete public build log contains only:

  scheduling build on Metal builder "builder-nxrjjq"

• The structured public error contains only:

  Failed to build an image. Please check the build logs for more details.

• No evidence appears for:

  • Dockerfile parsing
  • FROM/base-image resolution
  • dependency installation
  • Worker compilation
  • pnpm deploy
  • final image assembly
  • image registry push
  • container creation
  • runtime startup
  • healthcheck execution

• Multiple Worker deployments at different commits and times stopped after assignment to the same Metal builder.

• A separate Migration service assigned to another Metal builder emitted normal BuildKit steps and completed successfully.

• The resolved Worker deployment manifest correctly selects:

  • Dockerfile builder
  • Dockerfile.worker
  • repository-root build context
  • no custom build command
  • no custom start command
  • /health healthcheck
  • one Singapore replica
  • ON_FAILURE with three retries

• Local Worker production build and deploy-tree validation pass, including:

  • dist/index.js production output
  • production dependencies
  • Node syntax and dependency resolution
  • non-root UID/GID 10001 runtime contract
  • /health and /ready route contracts

This does not appear to be an application-level build failure because no Dockerfile instruction is reached.

Could you please:

1. Inspect the internal Metal builder/allocation failure for the deployment IDs and UTC timestamps above.
2. Confirm whether builder-nxrjjq is unhealthy or has stale service-specific build state.
3. Reassign a future build to another healthy builder, or advise how to clear the affected service/builder state without changing application source.
4. Provide the hidden structured failure reason or internal error code, since the public logs contain no Dockerfile-level error.
5. Confirm whether it is safe to create one new deployment after any platform-side correction.

We have stopped further retries and have not modified source or Worker configuration after reproducing the platform-side failure.

No credentials, secrets, connection URLs, customer content, or personal data are included in this report.

Thank you.