[arun279]

The failure

#195's run 39862150429 failed native-e2e-ios-light (discovery) (job 120115822649) before any tap. Maestro's deviceInfo raised SocketTimeoutException: Connect timed out, followed by Device became unreachable. The artifacts show:

• The runner and the tap proxy started and answered. maestro-runner.log has deviceInfo: 200 and then deviceInfo through the tap proxy: 200 at 22:11:30. Maestro's status check, deviceInfo, setPermissions and terminateApp all reached the runner through the proxy between 22:11:56 and 22:11:57. So it is neither a start-up race nor the driver dying at start-up. maestro-runner.mjs reports ready only after both answer.

• After 22:12:02 no request reached the runner, but nothing had died. Maestro started assertVisible at 22:12:04.9, and its first deviceInfo gave up at 22:12:49.9 after 45 s of connect timeouts. The runner's log has no connection after 22:12:02. The process list taken after the flow still shows the proxy's node (7593), its xcodebuild (7802), the runner (8433) and Cue (10576). The proxy logged no unreachable-runner error, so the requests never reached it either.

• The host started swapping hard at that moment.

  	Swap used	Swapouts
  Before the flow	0	0
  After the flow	536 MB	35,828

  Lane health logged 1,239 MB paged in during the flow. The proxy's node process sat at 14 MB resident, mostly paged out.

• PosterBoard ran a wallpaper gallery refresh just before. At 22:12:01 it started pushUpdatedPosterDescriptors ("Push To Proactive"), and about ten wallpaper extensions came up: Gradient, Mercury, Kaleidoscope, Aegir, Unity, Rhizome, Weather, Pride, Extragalactic. Lane health's process census shows PosterBoard at 193 MB, MercuryPosterExtension at 161 MB, and so on. The wallpaper processes wrote 29,305 of the run's 57,910 simulator log lines. The discovery lanes in Deduplicate CI setup and drop stale configuration #193's run 39855874496 and Prune tautological, outmoded and duplicated tests #192's run 39850551247 logged no gallery refresh, and 24 and 284 wallpaper lines.

I could not prove from the artifacts how the swap turned into connect timeouts, as opposed to slow replies. The timing and the 29,305 against 24 or 284 wallpaper lines point at the gallery refresh, which is simulator behaviour rather than ours.

Change

So the next lane that stalls this way says so in its health report:

• Lane health reports how far swap grew during the flow (sysctl vm.swapusage before and after) and how many wallpaper gallery refreshes PosterBoard ran, recording the first five refresh lines.
• A real bug in lane health itself: its SpringBoard spindump never ran on this runtime. The run's health log says "could not find SpringBoard to sample", because the lookup assumed the process path has no space, and the iOS 26.4 runtime's path contains iOS 26.4.simruntime. The lookup now matches the whole path.

A possible mitigation for later is turning off PosterBoard's gallery refresh the way ios-disabled-daemons.tsv turns off other daemons. I didn't do it here: SpringBoard depends on PosterBoard for the home and lock screen, and this one failure isn't evidence enough to justify the risk.

Tests

• New unit tests using lines from this run:
  • swap in use parsed from sysctl;
  • one refresh counted per gallery refresh, ignoring that refresh's other lines;
  • SpringBoard found under the real iOS 26.4.simruntime path.
• The end-to-end lane-health test expects the two new lines.

Mutation check, each applied alone:

Break	Caught by
The old SpringBoard pattern	the SpringBoard test
Counting every refresh line	the refresh test
Reading total swap instead of used	the swap test and the end-to-end test
Dropping the refresh line	the end-to-end test

Footprint

No product code changes. CI and tooling: scripts/lane-health.mjs gains about 20 lines. Configuration: one cspell word. Tests: about 30 lines.

pnpm check passes on Node 22.23.4.

[github-actions]

Pull request footprint

measurement	base to head
Product code lines in core/src, native/src, native/app, and native/modules	+0
Test lines in test, packages/*/{test,__tests__,e2e}, and .maestro	+29
CI and tooling lines in .github and scripts	+22
Product comment lines identified by a comment prefix	+0

User-delivered artifact sizes

measurement	base	head	delta
Expo iOS JavaScript bundle, raw file	5.45 MB	5.45 MB	0 B
Expo Android JavaScript bundle, raw file	5.48 MB	5.48 MB	0 B
Firebase tester APK, arm64-v8a and all densities	41.16 MB	41.16 MB	0 B
Play download estimate, XXXHDPI arm64-v8a English Android 15	19.83 MB	19.83 MB	0 B

Most complex product function

measurement	base	head
Cyclomatic complexity, oxlint complexity	10	10
Cognitive complexity, Biome noExcessiveCognitiveComplexity	15	15