Repository navigation
iOS smoke lane: two untracked failure signatures from the #2491 attribution pass (MAIN_THREAD_TIMEOUT on alert dismiss; replay timeout_cleanup_pending) #3337
Description
Activity
Adding a third scenario-level signature per the #3336 coordinator pass (coordinating from #3343, which this supersedes for this signature).
3.
smoke:webview-remote-content:wait text "Jump to form" 20000→wait_deadline_exceededwith highreadableCaptures— run 37843633972 (PR #3336 lane,98321a279, 2026-10-08T21:16:46Z). Typed shape:timeoutMs: 20000,captures: 25,readableCaptures: 24,captureTruncated: true,diagnosticId: mv01do8x-5064e742. Every pollreadableat 250-900 ms — the route answered the whole budget; the WKWebView page simply was not in the tree. The post-failure snapshot seconds later (test/artifacts/ios-simulator/smoke/1791493983644-15998/failed-step-102-snapshot.json) already containsJump to formx2,Account form,Email addressunder ahealthy/treeverdict — the cold-WebContent load finished just past the deadline.Lane-corroboration (not mine): the scenario predates this PR (#3084/#2902/#2493 on main); main was green on iOS the same day (run 37815426076); #3330 and #3334 pass Smoke on the same simulator UDID (5FEB61C5-…) and runner class with no cross-ref concurrency group, so host sharing is in play. The retry policy added in #3336 declined to re-issue this by design (
wait_deadline_exceededis named non-retriable: a readable miss is the product answering).Decision rule for future occurrences (per the coordinator pass): this is a cold-host timing flake UNLESS it reproduces on
main, orreadableCapturesstays high across N runs with the page link never appearing in any capture — that second shape would make it a page-load-path product defect. The owning fix for a real latency class is thePAGE_LOAD_WAIT_MSbudget or the page-load path, never widening the lane's retriable-reason set.Signature 3 non-reproduction: re-run of 37843633972 @ 98321a2 (identical head, same simulator UDID) passed smoke:webview-remote-content — all checks green. Per the recorded decision rule this occurrence is the cold-host timing-flake shape: no main reproduction, and the original run's post-failure snapshot proved the content arrives seconds late. Keeping the signature filed here so a recurrence is matched rather than re-triaged.
Why
While attributing recent
ios.ymlmain failures for #2491 (see #2491 (comment)), two signatures had no owning issue:MAIN_THREAD_TIMEOUTonalert dismiss— run 37660079219,smoke:automation-inputscenario, step "dismiss native alert". Typed details:reason: runner_main_thread_timeout,dispatched: "unknown",runnerFailureReason: runner_main_thread_execution_timeout, hint says the runner abandoned the command's main-thread work past its execution watchdog. The finished-at-boundary misreport fix(ios-runner): runMainThreadWork reports MAIN_THREAD_TIMEOUT for work that finished at the timeout boundary #2782 is closed and does not cover this shape; looks like a genuine main-thread stall during alert dismissal. Adjacent prior art: iOS alert activation can tap after the command deadline #2956, iOS runner smoke: alert-observation XCTests are red or flaky on main #2546.TIMEOUTwithtimeout_cleanup_pending— run 36115336398, failing step "Run gesture pan-duration smoke replay".examples/test-app/replays/gesture-pan-duration.ad, attempt 1/3: awaitcommand hitTIMEOUT after 180000ms(timeoutMode: cooperative,timeoutCleanupPending: true, junit wall 182s), and the attempt's runner log ends in** BUILD INTERRUPTED **. The settings replay in the same run passed. A command timeout leaving cleanup pending is its own class — the retry policy in fix(test): make the iOS simulator smoke lane's red/green signal meaningful (#2491) #3336 deliberately does not re-issue hard timeouts.Evidence
Both runs upload a single artifact named
ios-artifacts:gh run download 37660079219 --repo callstack/agent-device -n ios-artifacts→work/agent-device/agent-device/test/artifacts/ios-simulator/smoke/1791396322253-59034/failed-step.txt(plusstep-history.json,failed-step-55.png, snapshot)gh run download 36115336398 --repo callstack/agent-device -n ios-artifacts→work/agent-device/agent-device/test/artifacts/replays-ios-gesture-pan-duration/3be9273812dd9d58/examples__test-app__replays__gesture-pan-duration.ad/attempt-1/failure.txt(plusreplay-timing.ndjson; runner log under.tmp/agent-device-state/sessions/…gesture-pan-duration_attempt-1/runner.log)Scope
Triage each signature to an owning defect (runner main-thread stall vs alert-dialog path for #1; timeout-cleanup lifecycle / cooperative-timeout teardown for #2), then make each red step assert its real contract. No retry-layer change is implied — #3336's policy intentionally leaves hard timeouts red.