CI: Bazel caching, superseded-run cancellation, checkout@v7 - #32
Merged
Merged
Conversation
Every job cold-built until now — each leg re-fetched external deps and recompiled the runtime, googletest, and (benchmarks) BoringSSL + asio from scratch on every push. Add bazel-contrib/setup-bazel to each Bazel-running job with the bazelisk binary cache, the repository cache (external module downloads), and a per-config disk cache (action results), so branch builds restore from the caches main builds save. The codegen job already caches through setup-gradle; lint runs no Bazel. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un
Match the current release (as used by muchq/MoonBase's branch builds). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un
The runner image ships packages.microsoft.com repos that intermittently fail signature checks and abort apt-get update; libboost-dev comes from the Ubuntu archive, so let update fail soft and let install be the arbiter. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un
- concurrency: a new push to a PR cancels the in-flight run for the old head; main builds keep a per-run group so every merge still populates the bazel caches. - actions/checkout v4 -> v7, clearing the Node 20 deprecation warnings on every job. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un
aaylward
changed the title
gcc sizes the snprintf against full int widths (73 bytes worst case); give the buffer headroom for that instead of the real bounded output. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un
The asan CI leg caught //runtime:beast_transport_test dying with "Broken pipe": in RejectsOversizedBodies the Beast server closes the connection while SocketHttpClient is still streaming the request body, and the client's next send() raised SIGPIPE, killing the process. The race only fires when the close beats the remaining writes (sanitizer slowdown made the server lose it), which is why the test usually passes. Suppress the signal at the transport layer so a peer disconnect always surfaces as a write error: MSG_NOSIGNAL on send (Linux), SO_NOSIGPIPE on the socket (macOS/BSD, which lack MSG_NOSIGNAL); Windows has no SIGPIPE. Regression test drives a 64 MiB+ body into SocketHttpServer, whose content-length rejection closes mid-send the same way. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un
4 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
CI reliability/speed pass, patterned on muchq/MoonBase's branch builds:
Bazel caching — every job cold-built until now: each leg re-fetched all external modules and recompiled the runtime, googletest, and (benchmarks) all of BoringSSL + asio on every push. Adds
bazel-contrib/setup-bazel@0.19.0(version matching MoonBase) to each Bazel-running job with three caches:linux-gcc,asan-ubsan,benchmarks-opt, …) since object files differ across compilers/configs. PR branches restore the caches main builds save, so a typical PR leg recompiles only what its diff dirties.Not touched: the codegen job (already cached via
setup-gradle) and lint (runs no Bazel). The action wires cache paths through the user-level bazelrc, so existingbazeliskinvocations pick them up unchanged.Superseded-run cancellation —
concurrencygroup per ref withcancel-in-progress: true: a new push to a PR cancels the in-flight run for the old head. main builds get a per-run group (github.run_id) so every merge still completes and populates the caches.actions/checkoutv4 → v7 — clears the Node 20 deprecation warning emitted on every job.lint hardening (picked up while iterating here): the runner image's
packages.microsoft.comapt repos intermittently fail signature checks and abortedapt-get update;libboost-devcomes from the Ubuntu archive, so update now fails soft and the install step is the arbiter.Known trade-offs: GitHub's 10 GB per-repo cache quota is shared across the disk caches (eviction means a slower run, not a failure), and toolchain-change PRs see near-cold builds since their diffs invalidate most action results.
Testing
--configflags and test commands untouched; YAML-only.Checklist
🤖 Generated with Claude Code
https://claude.ai/code/session_019ytv3VMrURFYP2mdWhk3un