try GLOO_SOCKET_IFNAME: "eth0"

Borda · Borda · commit 4658409b399e · 2025-06-17T11:45:43.000+02:00
diff --git a/.github/workflows/ci-tests-fabric.yml b/.github/workflows/ci-tests-fabric.yml
@@ -80,6 +80,7 @@ jobs:
       TORCH_URL_TEST: "https://download.pytorch.org/whl/test/cpu/"
       # TODO: Remove this - Enable running MPS tests on this platform
       DISABLE_MPS: ${{ matrix.os == 'macOS-14' && '1' || '0' }}
+      GLOO_SOCKET_IFNAME: "eth0"  # trying to avoid "gloo" issue with SIGABRT
     steps:
       - uses: actions/checkout@v4
 
diff --git a/tests/tests_fabric/utilities/test_distributed.py b/tests/tests_fabric/utilities/test_distributed.py
@@ -123,7 +123,6 @@ def _test_all_reduce(strategy):
         [torch.device("cpu"), torch.device("cpu")],
     ],
 )
-@pytest.mark.flaky(reruns=3)  # flaky with "process 0 terminated with signal SIGABRT" (GLOO)
 def test_collective_operations(devices, process):
     spawn_launch(process, devices)
 

Original file line number	Diff line number	Diff line change
`@@ -123,7 +123,6 @@ def _test_all_reduce(strategy):`
`123`	`123`	`[torch.device("cpu"), torch.device("cpu")],`
`124`	`124`	`],`
`125`	`125`	`)`
`126`		`-@pytest.mark.flaky(reruns=3) # flaky with "process 0 terminated with signal SIGABRT" (GLOO)`
`127`	`126`	`def test_collective_operations(devices, process):`
`128`	`127`	`spawn_launch(process, devices)`
`129`	`128`