Repository navigation
Windows source CI intermittently fails the log batch off test #2012
Description
The Windows source job intermittently fails test_log_batch.py in the off case on unmodified main. The test requires output within 200 ms of enqueue, which includes scheduling both the writer and the thread that observes its output.
- Main at
9447e1deand main atabd5e5f0failed the same assertion. - Main at
ab0aee50passed with the same log writer and fixture. - On Windows with LLVM, delaying the writer's startup by 350 ms reproduced the same assertion failure. The other four cases passed.
I'll make the off case observe whether the extracted writer enters a finite batching wait, while keeping the exact-output assertion and bounded wait for completion. Scope is Windows source CI. S3, affected runner share unknown, confidence 1.0 from local reproduction. Discord claim.
Activity
NotRexed commentedon Nov 11, 2026
Fixed on main by #2013 (merged): the
offcase now times three deliveries and checks that the fastest is under 200 ms. Your repro, the writer's startup delayed 350 ms, only delays the first delivery, so the fastest of the other two still tellsofffrom the default's 250 ms batch, and a real batching regression still fails all three. Thanks for the analysis. If it flakes again on main after #2013, reopen this and your observe-the-wait approach is welcome.