Repository navigation
mcpx: mcpclient stress suite hits rotating pre-existing DATA RACEs (legacy reinit + handoff tests) #2842
Description
Found by the queue worker while shepherding PR #2760 (issue #2079) — the stress: PR changed tests (race/shuffle/repeat) job fails with a DIFFERENT pre-existing DATA RACE on successive runs, in legacy-session / handoff code that the PR under test does not touch:
- Run 41129819201 attempt 1 (PR feat(mcpx): handle upstream -32042 in CallTool — broker URL elicitations, retry once (#2079) #2760 head f3cd691):
TestLegacySession404Recovery/concurrent-calls-share-one-reinitialize— races betweenmirrorsHeaders()(modern.go:600) /callReissue()(client.go:1068) reads inCallToolandfinishLegacy()writes (client.go:460) during concurrent re-initialization. - Same run, rerun attempt:
TestClientHandoffCommitsOnlyWhenActivatedDATA RACE — also independently observed by another worker lane on unrelated run 41112180105. - Earlier run 41121731190 on the same PR:
TestDetachRefusesWhileACallIsInFlightDATA RACE (handoff Detach vs stdio recvLoop).
The same PR head passed the stress job outright on run 41121731190 (mcpx workflow success, head 3adca52), and each rerun fails on a different test — the signature of a flaky race family in the legacy-reinit/handoff surface, not a regression in any one PR.
Acceptance:
- Reproduce each race locally with
go test -race -count=N -shuffle=onon the named tests against clean main. - Fix the unsynchronized field access (the
finishLegacywrites vs the header-mirror / reissue reads are the concrete pair from run 41129819201) or serialize the concurrent-reinit path. - Stress job green on repeated runs for the affected tests.
Filed as the pre-existing-failure record for PR #2760 merging under the standing criteria (its own tests verified; Codex clean on head f3cd691).
Acceptance criteria
- Each named race (legacy reinit mirrorsHeaders/callReissue vs finishLegacy, TestClientHandoffCommitsOnlyWhenActivated, TestDetachRefusesWhileACallIsInFlight) is reproduced on clean main with
go test -race -count=N -shuffle=on, or documented as no longer reproducible with the runs cited - The unsynchronized field access in the legacy-reinit/handoff surface is fixed (or the concurrent-reinit path is serialized) with the race tests passing repeatedly under -race
- The mcpx stress job is green on repeated runs for the affected tests — no rotating DATA RACE failures across reruns
Metadata
Metadata
Assignees
Labels
Package/app: mcpx (pkgs/ or apps/ in this repo)Medium: expected under 1 day of workSomething isn't working (prefixed form of the classic bug label)Intake review done: acceptance criteria, links, and normalized labels addedOrdinary priority (the default)Filed or worked by the queue/worker pilotQueue packet kind: verify, reproduce, and route this item
Activity
dezren39 commentedon Oct 30, 2026
Intake review
Fleet monitor intake pass (protocol §5). Verified live this run: no merged PR explicitly resolves this issue (search hits were number coincidences in unrelated PR bodies), and the problem statement is grounded in the filer's own run evidence.
Related: Related: PR #2760 / issue #2079 (shepherded when found — the PR itself passed the stress job on one head; the races are in code it does not touch).
Notes / risks / suggested approach: A rotating race family is worse than a single flaky test: it taxes every mcpx PR's stress job with reruns and can mask a real regression. Risk of the fix itself: medium — concurrency changes in legacy-session/handoff code need the -race suite run repeatedly. Suggested approach: reproduce per-test with -race -count on clean main first, fix the finishLegacy write/read pair, then the handoff pair.
dezren39 commentedon Oct 31, 2026
Observation from queue task 3108 (lane C, 2026-10-31): PR #2975 stress run 41198959299 also failed internal/daemon TestRegistryHandoffKeepsTheChildAcrossTheSuccessor (0.31s) alongside the mcpclient handoff race in this issue. The daemon test passed 5/5 with -race on the PR branch on the laptop, and TestClientHandoffCommitsOnlyWhenActivated reproduced its DATA RACE on clean main d4279a4 (2/3 count=10 race/shuffle runs) as well as on the branch — same pre-existing family, not a PR regression. Adding the daemon registry-handoff test name as another member of this family.
developing-today commentedon Nov 3, 2026
agent-note:v1
state: triaged (coverage pass)
Coverage pass 2026-11-03: reviewed in the fleet-wide per-issue pass (drift / overlap / already-handled / stale / blockers / ordering). Verdict: still live; queued in repo-health-code-pilot as triage at priority 40 (priority/normal). If this is already handled at current main, the triage closes it with evidence; if it needs splitting, the triage files the split.
Found via: coverage pass 2026-11-03
38 remaining items
dezren39 commentedon Nov 3, 2026
🤖 Agent note · parked
dezren39 commentedon Nov 3, 2026
CI backlog tracked in issue #5194 (basic-tier Actions runs queued 30+ min without starting). The park on this issue's PR resumes once those runs start and pass.
Merge pull request #5139 from developing-today/queue/8677-2842-mcpcli…
dezren39 commentedon Nov 3, 2026
agent-note:v1
state: done
agent: repo-worker-code
task: 8677
lane: code-n
at: 2026-11-03T22:53:30Z
Resumed task 8677 from its awaiting_ci checkpoint and finished it. Fresh verification: PR #5139 checks all green (basic: mcpx fix carries a test — pass; basic: production line limit (2500) — pass; CodeRabbit skipping/neutral). Codex gate: latest Codex review on head 9026fba reports no findings / no unresolved findings; the earlier P1 (detached-read regression coverage) was dispositioned in that head with TestRecvQueuedBehindDetachDoesNotConsumeSuccessorFrame plus a thread reply.
Merged PR #5139 (merge commit bc764d4); issue #2842 auto-closed. Completing task 8677 as fixed_merged.