Skip to content

Without libSkiaSharp, every app publish fails, and the container still goes healthy #219

Closed

Description

@SShadowS

Follow-up to #205, seen on BC 29.0.54011.57673 with an older bc-runner image (digest sha256:1724098..., from before #201/#212).

That image has no native for SkiaSharp 4.152.1, so the entrypoint logs the WARN and links neither Skia nor harfbuzz. The WARN says report rendering stays disabled. In practice it is worse: every publish through the dev endpoint fails, also apps with no reports at all (Any, Library Assert, Test Runner):

The type initializer for 'SkiaSharp.SKAbstractManagedStream' threw an exception.
The type initializer for 'SkiaSharp.SkiaApi' threw an exception.
Unable to load shared library 'libSkiaSharp' or one of its dependencies.
/bc/service/libSkiaSharp.so: cannot open shared object file: No such file or directory

The default-flow test toolkit publish only echoes the HTTP code, so the boot carries on, /tmp/bc-ready is written, the healthcheck passes, and the database has no Test Runner, Library Assert or Any. I found out when an agent asked me why its test runs had no toolkit.

With the image I have, I fixed it by hand: the 4.152.1 linux-x64 native from NuGet into /bc/service/libSkiaSharp.so, the harfbuzz symlink, restart-service.sh (the failed type initializer stays cached until NST restarts), then the toolkit published fine (8 apps, all HTTP 200).

#212 covers the normal case now. What is left is the case where that NuGet download fails (offline host, proxy, nuget.org down):

  • The WARN could say that publishing is broken too, not just reports.
  • The toolkit publish could fail the boot (or at least not touch /tmp/bc-ready) when an app comes back non-2xx, so the healthcheck tells the truth.

Happy to send a PR for the second point if you want it.

Activity

  1. StefanMaron commented on Nov 7, 2026

    @StefanMaron
    Collaborator

    Fixed on master by #220, commit 3321941. Thanks for the clear report. I did the second point myself, so no PR is needed.

    • The keep-set install, the default test toolkit publish and the Test Runner Extension publish now record a failed publish and print the response body. If any failed, the entrypoint lists them, stops NST, and the container exits with rc 1. docker compose up --wait and wait-for-bc-healthy.sh then fail right away. BC_ALLOW_PUBLISH_FAILURES=1 finishes the boot anyway.
    • The SkiaSharp warning now says app publishing can fail too, not only report rendering.

    I reproduced your case on BC 29.0 by removing the native from the image and making nuget.org unreachable. I got the same SkiaSharp.SkiaApi error, and the container exited after 91 s, listing the 8 apps. On BC 27.5 with SkiaSharp 3.119.0, the same setup still published every app. That is why the warning says "can fail" and not "will fail".

    The image with this change is built from master by build-image.yml. Please reopen if you still see a healthy container with a missing toolkit.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions