Repository navigation
test: GraalPy experimental leg loses an xdist worker and fails tests on main #2082
Description
Found in the CI reports. Discussion #2072 reports Experimental run 39898991163 on main at 61bb4ba66.
- The warnings:
Python graalpy-25.4 (experimental)warnedtests failed on Python graalpy-25.4 (experimental, non-blocking)andtests lost an xdist worker, and its job summary says steptestsfailed. - What it means: the leg stays green by design (non-blocking), but a lost xdist worker usually means a crash, OOM or hang, and the failing ids are unknown.
- Next: a worker is reading the log to find which shard and tests failed, whether it's new on GraalPy or recurring, and the cause.
Activity
JarryShaw commentedon Oct 23, 2026
It recurred on
mainat221e9da74(Experimental run 39910421128, job 120272075198; Discussion #2072). The GraalPy leg's only failure wastest_module_isolation.py::OrderIndependenceTests::test_the_reverse_order_passes_too, whose nested pytest hitTimeoutExpired … timed out after 600 seconds. The last shard took 1033 s (1 failed, 482 passed). #2086 fixes this and is ready to merge. The run's other shards passed, and no worker went down.test(conftest): skip the tier-guard lock under an ancestor's shared h…
JarryShaw commentedon Oct 23, 2026
All three parts are now handled:
test_module_isolation's nested pytest queued behind a writer at test: the tier guard's blocking flock deadlocks GraalPy's xdist session #1977's gate, is fixed by test(conftest): skip the tier-guard lock under an ancestor's shared hold (#2082) #2086, merged asbfa2469cc.import *premise test is fixed in test: GraalPy's 'from m import *' swallows a missing __all__ name, failing the aenum premise test #2083.