Repository navigation
Conversation
…aits Both Windows rendezvous tests poll for the startup thread to be observed holding cbm-startup-v2.lock. That state is stable once reached — the test's own reader lock blocks the rendezvous write, so the startup lock stays held — only the onset (thread spawn + IPC preamble) is timing-dependent, and the 200ms budget missed it on a starved x64 runner (release run 30322785509, test_daemon_ipc.c:1075 ASSERT(startup_observed), first occurrence in 6 runs). Raise both onset budgets to 10s; a healthy run still exits on first observation. Per the flaky-test ladder this is the legitimate budget case: stable awaited state, budget as the sole flake source. Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
|
Closing unmerged: falsified by local verification before merge. With the 10s budget in place, 2× ARM64 VM runs still failed the sibling test ( |
1 participant
Release run 30322785509 failed on windows-latest x64 1/2:
test_daemon_ipc.c:1075 ASSERT(startup_observed)— first occurrence in 6 runs. Attribution: the awaited state (startup thread holdingcbm-startup-v2.lockwhile the test's reader lock blocks the rendezvous write) is stable once reached; only the onset is timing-dependent, and the 200ms poll budget missed thread-spawn + IPC preamble on a starved runner. Per the flaky-test ladder this is the legitimate budget-bump case (stable state, budget sole flake source): both sibling onset waits go 200ms → 10s, healthy runs still exit on first observation. 46/46 locally; 2× ARM64 VM verification in flight. Unblocks v0.9.1-rc.1.