Repository navigation
test(relay): a failed drill names a shaper that stopped forwarding - #5935
Conversation
Path's Drop printed only the shaper's counters, so a shaper that stopped mid-drill surfaced only as a client status timeout. It now prints the shaper's verify verdict, which leads with why forwarding stopped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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
WalkthroughWhen a drill path is dropped with a remaining shaper, Priority: ⬇️ Low Merge Risk: ⚪ Minimal · up to
|
|
Merge summary: on a drill failure, Decisions: the maintainer accepted reusing Reviews: Codex (no findings, plus thumbs-up) and CodeRabbit (no actionable comments) both cover head b85ea47. CI is green. Enqueuing. (Written by Claude Opus 5.5) |
Quest:
quest/m1/test-flakes-2/drill-shaper-failure.md(deleted here, and removed from the epic'sRequired).Problem
When a drill's shaper stops forwarding mid-run, its clients just time out.
Drop for Pathprinted only the shaper's counters, and onlyPath::verify()(called at the very end of a drill) reported the shaper's error. So the CI log showedsubscriber: no status change within 10s, wanted Connectedand nothing else. That is how #5796 hid.Fix
Drop for Pathnow printsShaper::verify()'s verdict instead of bare counters. Every drill callsPath::verify()at its end, which takes the shaper, so thisDropbranch only runs when a drill fails first. When the shaper stopped, the verdict leads withthe shaper stopped forwarding: <cause>. Otherwise it prints the same counters as before, plus any impairment that never acted.Reproduction
I temporarily made the shaper's reply task fail after 20 datagrams and ran
relay_killed_mid_group_aborts_then_resumes::loopback:shaper: up: 62 packets, ...then the panicsubscriber: no status change within 10s, wanted Connected. Nothing names the shaper.shaper: the shaper stopped forwarding: sabotaged for the repro (up: 63 packets, ...)before the same panic.Regression test
None landed. Testing this would mean a test-only hook in
moq-shaperto force a failure, plus capturing stdout from aDrop. That is a lot of new surface for a diagnostic print. The repro above is the proof.Choice made without asking
Dropreusesverify()instead of adding an accessor tomoq-shaperthat returns only the failure. Reusing it adds no API. The cost: a drill that fails early may also print that some impairment never acted. That is true, and labeled as such, but it is not the reason the drill failed.Checks
just check: pass (580 tests, including all 14 drills)quest check: passAPI / wire impact
None. Test-only change in
rs/moq-relay/tests/drills.rs.(Written by Claude Opus 5.5)
🤖 Generated with Claude Code