Follow-up from PR #638 review (#638 (comment)).
marketplace-release.yml invokes the reusable marketplace-build.yml without build_cuda, which defaults to true, so build-marketplace-cuda (runs-on: [self-hosted, linux, x64, gpu]) always runs on release. continue-on-error: true handles step failures gracefully, but if no runner with those labels is registered the job stays queued indefinitely and downstream jobs (needs: [..., build-marketplace-cuda]) stall the whole release.
Possible mitigations:
- add a
timeout-minutes to the CUDA job so a missing runner fails fast instead of queueing forever
- split the CUDA variant build into its own non-blocking workflow
- operational: ensure the GPU runner stays registered, or dispatch with
build_cuda=false
Follow-up from PR #638 review (#638 (comment)).
marketplace-release.ymlinvokes the reusablemarketplace-build.ymlwithoutbuild_cuda, which defaults totrue, sobuild-marketplace-cuda(runs-on: [self-hosted, linux, x64, gpu]) always runs on release.continue-on-error: truehandles step failures gracefully, but if no runner with those labels is registered the job stays queued indefinitely and downstream jobs (needs: [..., build-marketplace-cuda]) stall the whole release.Possible mitigations:
timeout-minutesto the CUDA job so a missing runner fails fast instead of queueing foreverbuild_cuda=false