Persistent build failures on Metal builder (builder-ppcgmd)

eljimara

FREEOP

7 days ago

Subject: Persistent build failures on Metal builder (builder-ppcgmd) — "Failed to read app source directory", GitHub auto-deploy also not triggering

Project: khushu-tracker

Project ID: 624922aa-788b-4213-9e82-0b64730bb5a0

Service: khushu-tracker

Service ID: 7414bc25-6b55-4f84-9365-77088c995c92

Environment: production (c910efbf-64d5-443d-91a2-db736b3e7a5c)

Hi,

I've been trying to deploy a new commit to my khushu-tracker service for about an hour and every attempt fails at the build stage.

What I'm seeing:

Every railway up from the CLI fails with the same error, at the same step, every time:

scheduling build on Metal builder "builder-ppcgmd"
fetched snapshot sha256:05784f32b481862b72fd13623b0f5fa5758ece1dbc0df49a2059bd129741bffe (161 kB bytes)
[INFO] fetching snapshot
[INFO] unpacking archive
using build driver nixpacks-v1.41.0
Error: Failed to read app source directory

Caused by:
    No such file or directory (os error 2)
nixpacks exited with an error

This has happened on 5 separate attempts over roughly an hour, always on the same builder (builder-ppcgmd).

Also affected: GitHub auto-deploy is enabled (Settings → Source shows "Auto deploys when pushed to GitHub" is ON, repo Eljimara/khushu-tracker connected, root directory backend, branch main) but no deployment was triggered by a push to main about an hour ago — nothing shows up in the deployment history for that push at all.

What I've already checked/ruled out:

Verified the service's Source settings (repo, root directory, branch) are all correct.
Confirmed via railway.com/status that there's no publicly listed incident.
Used Railway's own "Diagnose" feature on the failed deployment — it returned:

Diagnosis > Infrastructure Error: "Try redeploying this service, as the build failure looks like a transient builder issue rather than a problem with your code or config. The build failed at the snapshot unpacking stage with 'Failed to read app source directory: No such file or directory', even though the same commit and railway.json had built successfully just an hour earlier."
Retried after receiving that diagnosis — same failure, same builder, same error.
The last successful build for this service was deployment 4e20dd84-ad7e-42e0-9cfa-ba84b8a58304 (2026-08-14 15:33 UTC), using the exact same railway.json / Nixpacks setup that's now failing.

Could you take a look at builder builder-ppcgmd for this service, and/or find out why the GitHub webhook isn't triggering an auto-deploy? Happy to provide any of the failed deployment IDs if useful — a few of them:

c3d53966-3fbd-49c1-b5d9-d0f977009d26
86d2791c-24bf-4f2d-8eb4-b31d1d7e5369
b954ce34-2c0e-4eed-9103-bd73a1235f8d
6dfcea96-91ac-48d7-81ef-3d55c8a46384
b704a8d4-4714-4ff7-a3bf-d2ed8c3cef05

Thanks!

Solved$10 Bounty

Pinned Solution

mayori

MODERATOR

7 days ago

Try Deploy Latest Commit from the Command Palette

4 Replies

Railway

BOT

7 days ago

This thread has been opened as a bounty so the community can help solve it.

Status changed to Open Railway • 7 days ago

0x5b62656e5d

MODERATOR

7 days ago

Try migrating to Railpack or Dockerfile. Nixpacks is deprecated.

eljimara

FREEOP

7 days ago

Update: this isn't a Nixpacks-specific issue.

Following the moderator's suggestion, I switched the service from Nixpacks to Railpack, then to a plain Dockerfile — and all three fail, always on the same builder (builder-ppcgmd), always at snapshot unpacking, before the actual builder tooling really gets going:

Nixpacks:

Error: Failed to read app source directory
Caused by:
    No such file or directory (os error 2)

Railpack (different failure, but also right after unpacking — its own "mise" install step broke):

using build driver railpack-v0.36.4
railpack prepare exited with an error

Dockerfile (new deployment ID: 3985481f-c7b7-4d92-bc98-1cf882e9d2a9):

[INFO] unpacking archive
Build Failed: fsutil.NewFS(/var/lib/builder/build-sessions/3985481f-c7b7-4d92-bc98-1cf882e9d2a9/snapshot-target-unpack/backend): resolve : lstat /var/lib/builder/build-sessions/3985481f-c7b7-4d92-bc98-1cf882e9d2a9/snapshot-target-unpack/backend: no such file or directory

All three builders fail trying to read the source tree from the unpacked snapshot at the service's configured root directory (backend) — the Dockerfile error is the clearest: snapshot-target-unpack/backend genuinely doesn't exist after "unpacking archive" reports success. This looks like a bug in how builder-ppcgmd unpacks snapshots for this service when a root directory other than / is configured, independent of which build system consumes it afterward.

Also still true: GitHub auto-deploy hasn't triggered a single build since a push ~2.5 hours ago, despite Settings → Source showing it enabled and correctly configured (repo, root directory backend, branch main).

Happy to provide the new deployment ID or anything else that helps narrow it down.

eljimara

Update: this isn't a Nixpacks-specific issue. Following the moderator's suggestion, I switched the service from Nixpacks to Railpack, then to a plain Dockerfile — and all three fail, always on the same builder (`builder-ppcgmd`), always at snapshot unpacking, before the actual builder tooling really gets going: **Nixpacks:** ``` Error: Failed to read app source directory Caused by: No such file or directory (os error 2) ``` **Railpack** (different failure, but also right after unpacking — its own "mise" install step broke): ``` using build driver railpack-v0.36.4 railpack prepare exited with an error ``` **Dockerfile** (new deployment ID: 3985481f-c7b7-4d92-bc98-1cf882e9d2a9): ``` [INFO] unpacking archive Build Failed: fsutil.NewFS(/var/lib/builder/build-sessions/3985481f-c7b7-4d92-bc98-1cf882e9d2a9/snapshot-target-unpack/backend): resolve : lstat /var/lib/builder/build-sessions/3985481f-c7b7-4d92-bc98-1cf882e9d2a9/snapshot-target-unpack/backend: no such file or directory ``` All three builders fail trying to read the source tree from the unpacked snapshot at the service's configured root directory (`backend`) — the Dockerfile error is the clearest: `snapshot-target-unpack/backend` genuinely doesn't exist after "unpacking archive" reports success. This looks like a bug in how `builder-ppcgmd` unpacks snapshots for this service when a root directory other than `/` is configured, independent of which build system consumes it afterward. Also still true: GitHub auto-deploy hasn't triggered a single build since a push ~2.5 hours ago, despite Settings → Source showing it enabled and correctly configured (repo, root directory `backend`, branch `main`). Happy to provide the new deployment ID or anything else that helps narrow it down.

mayori

MODERATOR

7 days ago

Try Deploy Latest Commit from the Command Palette

mayori

Try Deploy Latest Commit from the Command Palette

eljimara

FREEOP

7 days ago

That worked — thank you! "Deploy Latest Commit" went through fine and the service is back online with the latest code. Looks like the bug was specific to the snapshot upload/unpack path used by railway up and the (apparently broken) GitHub webhook auto-deploy, not to deploying via GitHub through the dashboard.

Marking this resolved. Thanks for the quick help!

Status changed to Solved 0x5b62656e5d • 6 days ago

Welcome!