Repository navigation
test(drill): leak the session's sources too in subscriber-leaks-broadcasts - #5338
Conversation
…casts Since #5054 a minted source's guard lives in the driver's SourceServe, not in Announced, so leaking Announced alone left the source to close with the session and the front closed with it. Leak both the routes and the sources. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 22 minutes. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
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. |
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI)
Reviewed full head 5b85ffd. No actionable bugs found.
Direction: keeping this confined to the mutation is appropriate. Leaking both PrefixRun::announced and SourceServe::source removes the two independent cleanup paths the cancellation drill observes; changing the shared SourceGuard destructor would broaden the mutation unnecessarily.
Verification: inspected all four patch hunks, subscriber ownership/teardown, SourceGuard, the cancellation drill, and the sensitivity runner. All hunk counts and old-side text match the source at this SHA. This was a GitHub-only static review; I did not apply the patch, compile, or run tests. No existing substantive review findings were present to duplicate.
|
Merge summary: test-only change to the drill mutation. CI green; OpenAI and Codex reviewed the head (5b85ffd) with no findings. No changes made at merge time; enabling auto-merge on the current head. (Written by Claude Opus 5.5) |
Problem
Nightly
just test drill-sensitivityhas failed since 2026-10-09:cancel_under_backpressure_releases_the_reader::loopbackstill passed withsubscriber-leaks-broadcastsapplied. The cause is deterministic, not a flake: the drill passed in 0.4s under the mutation.#5054 moved each minted source's
SourceGuardout ofAnnouncedRoute.sources(insideAnnounced) and into the driver'sSourceServe. The patch still applied, but leakingAnnouncednow leaks only the routes. When the session ends,SourceServedrops and closes the source, and the reader's origin front closes with it.The late-September nightly failures were a different mutation (
relay-withdraws-lost-publisherbroke the build).Approach
The front closes when its route is retracted or when its source closes, and the session owns both. The mutation now wraps both in
ManuallyDrop:PrefixRun::announced(routes) andSourceServe::source(sources). Leaking just one isn't enough, because the other release still closes the front. I verified this locally: each half alone passes the drill, and both together fail it. The drill and the sensitivity check are unchanged.Local evidence:
the cancelled subscriber's broadcast never closed, consistently across repeated runs.sensitivity.sh --apply-only: all 6 mutations apply.just checkpasses.Impact
Alternatives
SourceGuard::dropa no-op: this would also hit the IETF subscriber and every other user of the guard, so it's broader than this behavior.Follow-ups
🤖 Generated with Claude Code
(Written by Opus 5.5)
Need help on this PR? Tag
@codesmith-botwith what you need. Autofix is disabled.