You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fix IPC memory pool leak in nightly repeated test runs (#1905)
* Fix IPC memory pool leak in nightly repeated test runs
IPC test fixtures and test methods create DeviceMemoryResource and
PinnedMemoryResource instances that are never closed. Under single-run
CI this is harmless, but nightly CI runs tests with pytest-repeat
--count=100, accumulating ~200+ leaked CUDA mempools and POSIX fd
exports per test file until the process segfaults around iteration 83.
Convert the ipc_memory_resource fixture from return to yield with
explicit close() teardown, and close locally-created memory resources
in test_errors, test_send_buffers, and test_workerpool after use.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Wrap IPC memory resource cleanup in try/finally
Ensure MR.close() runs even when tests fail (timeout, assertion,
queue error) so pools don't leak on the unhappy path either.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
0 commit comments