Summary
E2E - Flutter [iOS] is flaky on GitHub-Actions macOS runners due to a random, per-session CoreSimulator app-launch wedge, not a cold-start/warmth problem. It is quarantined (removed from the CI Status gate's needs in .github/workflows/ci.yml) as of #398/#387 — the job still runs and is visible on PRs but does not block. RN iOS stays gated, so iOS-mobile coverage isn't lost.
Symptom
On some runs, 1–3 specs fail at session-create, ~7 min each, with either:
WebDriverError: The operation was aborted due to timeout on POST /session (the wdio client aborts at connectionRetryTimeout), and in the appium log:
Command 'xcrun simctl launch --terminate-running-process <udid> com.facebook.WebDriverAgentRunner.xctrunner' timed out after 600000ms (also seen on simctl terminate), orXCTDaemonErrorDomain Code=5 "Timed out attempting to launch app."
appium-xcuitest launches the preinstalled WDA via simctl launch per session (appium-webdriveragent launchWithPreinstalledWDA). That simctl exec intermittently wedges the full 600 s node-simctl timeout; appium:wdaLaunchTimeout does not bound it and there is no cap to shorten it.
What was ruled out (PR #398/#387)
- usePreinstalledWDA (vs usePrebuiltWDA/xcodebuild): fixed RN iOS and removed the in-session compile, but Flutter iOS still per-session-launches the WDA xctrunner.
- Stripping embedded
Frameworks/XC*: a real-device-only step; on the simulator it makes the runner crash on launch (SBMainWorkspace denial). Removed.
skipLogCapture + appium:dartVmServicePort pin (to drop the concurrent simctl log stream): non-viable — the iOS Flutter engine ignores the VM-service port pin (documented in e2e/wdio.flutter.conf.ts), so syslog is the only iOS VM-discovery path and can't be disabled.- SpringBoard / WDA-launch warm-up, incl. a blocking gate that required a fast
simctl launch before grading: the gate confirmed "warm" (round 1) yet grading still wedged 2 specs — warmth does not predict graded-session success, confirming the wedge is random per-session.
specFileRetries (+ deferred): a session-create timeout kills the worker in a way WDIO does not re-run (verified: wedged spec reported failed with no extra worker run).
Proposed fix (the reason this is tracked, not abandoned)
External WDA: launch WebDriverAgent once in the workflow (replicating appium's exact env — USE_PORT + WDA_PRODUCT_BUNDLE_IDENTIFIER), retry until its /status answers, keep it resident, and attach every session via appium:webDriverAgentUrl. This eliminates the per-session simctl launch entirely, so the random wedge can't hit session-create. A first attempt failed on the standalone-WDA-serving detail (the manual simctl launch didn't bring up the :8100 server) — needs the env replicated correctly + a retry-until-serving loop.
Once external-WDA is reliably green across several cold-sim runs, re-add e2e-flutter-ios-macos to ci-status.needs to un-quarantine.
Scope note
The Flutter iOS package smoke runs inside the same job (after the e2e suite), so it is also non-gating while quarantined. RN iOS (both arches) + its package smoke, Flutter Android, and RN Android remain gated.
Summary
E2E - Flutter [iOS]is flaky on GitHub-Actions macOS runners due to a random, per-session CoreSimulator app-launch wedge, not a cold-start/warmth problem. It is quarantined (removed from theCI Statusgate'sneedsin.github/workflows/ci.yml) as of #398/#387 — the job still runs and is visible on PRs but does not block. RN iOS stays gated, so iOS-mobile coverage isn't lost.Symptom
On some runs, 1–3 specs fail at session-create, ~7 min each, with either:
WebDriverError: The operation was aborted due to timeoutonPOST /session(the wdio client aborts atconnectionRetryTimeout), and in the appium log:Command 'xcrun simctl launch --terminate-running-process <udid> com.facebook.WebDriverAgentRunner.xctrunner' timed out after 600000ms(also seen onsimctl terminate), orXCTDaemonErrorDomain Code=5 "Timed out attempting to launch app."appium-xcuitest launches the preinstalled WDA via
simctl launchper session (appium-webdriveragentlaunchWithPreinstalledWDA). Thatsimctlexec intermittently wedges the full 600 s node-simctl timeout;appium:wdaLaunchTimeoutdoes not bound it and there is no cap to shorten it.What was ruled out (PR #398/#387)
Frameworks/XC*: a real-device-only step; on the simulator it makes the runner crash on launch (SBMainWorkspacedenial). Removed.skipLogCapture+appium:dartVmServicePortpin (to drop the concurrentsimctl log stream): non-viable — the iOS Flutter engine ignores the VM-service port pin (documented ine2e/wdio.flutter.conf.ts), so syslog is the only iOS VM-discovery path and can't be disabled.simctl launchbefore grading: the gate confirmed "warm" (round 1) yet grading still wedged 2 specs — warmth does not predict graded-session success, confirming the wedge is random per-session.specFileRetries(+ deferred): a session-create timeout kills the worker in a way WDIO does not re-run (verified: wedged spec reported failed with no extra worker run).Proposed fix (the reason this is tracked, not abandoned)
External WDA: launch WebDriverAgent once in the workflow (replicating appium's exact env —
USE_PORT+WDA_PRODUCT_BUNDLE_IDENTIFIER), retry until its/statusanswers, keep it resident, and attach every session viaappium:webDriverAgentUrl. This eliminates the per-sessionsimctl launchentirely, so the random wedge can't hit session-create. A first attempt failed on the standalone-WDA-serving detail (the manualsimctl launchdidn't bring up the:8100server) — needs the env replicated correctly + a retry-until-serving loop.Once external-WDA is reliably green across several cold-sim runs, re-add
e2e-flutter-ios-macostoci-status.needsto un-quarantine.Scope note
The Flutter iOS package smoke runs inside the same job (after the e2e suite), so it is also non-gating while quarantined. RN iOS (both arches) + its package smoke, Flutter Android, and RN Android remain gated.