Repository navigation
quest(test): abandon the unreproduced auth serve signal flake - #5707
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
auth_serve_stops_on_sig{int,term} never failed again: not under load
averages of 100-175 on 24 cores, nor with both binaries evicted from the
page cache, on current main or on the flake-era b581c27. The slowest of
100 stressed runs took 0.39 s against the 10 s bound. The tests keep their
bounds and distinct expect messages, so a recurrence names its wait.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Outcome: the quest's first step (reproduce under load and record which wait expired) could not be completed. Across about 400 runs on current main and the flake-era (Written by Claude Opus 5.5) |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
WalkthroughThe auth-serve signals quest document and its entry in the Required list were removed. Priority: ⬇️ Low Merge Risk: ⚪ Minimal · up to
|
|
Merging. Decision confirmed by the maintainer: abandon the quest rather than keep it open, matching #5648 for the sibling export exit flake.
Suggested follow-up (not started): log (Written by Claude Opus 5.5) |
Summary
quest/m1/test-flakes-2/auth-serve-signals.mdas unreproducible and drop its line from thetest-flakes-2README. The flake is not resolved: which of the two 10 s waits expired on 2026-10-13 (test(net): pin cross-node close flapping and its residuals #5458, fix(hang): keep catalog wall clock exact to the nanosecond #5481) was never recorded.Reproduction attempts
All passed, 0 failures:
mainnextest --stress-count 100 --test-threads 24, load avg ~164 on 24 coresmainmoq-clinextest suite, load avg ~114mainPOSIX_FADV_DONTNEED) each run, load avg ~170b581c27ednextest --stress-count 100, load avg ~86b581c27edPhase timing of a bare
moq auth serveunder load: spawn to "auth server listening" 10-60 ms (100-160 ms cold), SIGTERM to exit 5-40 ms.Reading the code turned up no cause either. Since #5575,
Signals::new()installs both handlers beforeservebinds. Before it, thebiasedselect polledshutdown_signal()first and installed both before the "listening" log. The exit path is a select return plusshutdown_background(). Nothing in either interval waits on anything that CPU or page-cache pressure could stretch toward 10 s. Memory pressure (swap) is the remaining environmental candidate, which I did not induce on the shared host.The tests keep their bounds unchanged, and each wait has its own expect message (
auth server never listened/moq ignored the signal). So a recurrence names the wait that expired and can get a fresh quest with that evidence.Validation
quest check: passedjust check: not run (quest Markdown only, no code)Public API
None.
Wire impact
None.
Decisions
Suggested follow-ups
moq auth servelogs the configured--listenaddress, so--listen 127.0.0.1:0printsaddr=127.0.0.1:0instead of the bound port (rs/moq-cli/src/auth.rs,Serve::run). Logginglistener.local_addr()makes port 0 usable. Recommendation: a small [XS] quest, or fold it into the next auth serve change.(Written by Claude Opus 5.5)
🤖 Generated with Claude Code