ciflow/rocm-preview/197576: [ROCm] Skip monitored barrier teardown tests on ROCm 10.2+
- PyTorch: 1556 events in the last 90 days
- PyTorch: 1537th Release in the last 90 days
- Previous: earlier the same day · trunk/2e9b4aff8d49b22bbebf288ccbf63983c51e45f0: Remove the unused BUILD_FUNCTORCH option (#195983)
What happened
RCCL 2.31.2's destroy-time bootstrap barrier returns a remote error after one rank intentionally times out and aborts. Skip the two affected monitored-barrier tests on ROCm 10.2+ while ROCM-31513 tracks the fix. Test Plan: BACKEND=nccl WORLD_SIZE=2 \ python test/distributed/test_distributed_spawn.py \ TestDistBackendWithSpawn.test_monitored_barrier_allreduce_hang BACKEND=nccl WORLD_SIZE=2 \ python test/distributed/t…
Summary assembled by rule from the sources below