Issue 1165 : nccl + torch.cuda not available (#1166)

* fix issue 1165

* Update ignite/distributed/comp_models/native.py

Co-authored-by: vfdev <vfdev.5@gmail.com>

* add test for nccl /wo gpu

Co-authored-by: Desroziers <sylvain.desroziers@ifpen.fr>
Co-authored-by: vfdev <vfdev.5@gmail.com>

Author: 3 people

ignite/distributed/comp_models/native.py

@@ -74,6 +74,9 @@ def _create_from_backend(self, backend, timeout=None, **kwargs):
             if timeout is not None:
                 init_pg_kwargs["timeout"] = timeout
 
+            if backend == dist.Backend.NCCL and not torch.cuda.is_available():
+                raise RuntimeError("Nccl backend is required but no cuda capable devices")
+
             dist.init_process_group(backend, init_method="env://", **init_pg_kwargs)
             # https://github.com/facebookresearch/maskrcnn-benchmark/issues/172
             dist.barrier()

tests/ignite/conftest.py

@@ -216,3 +216,12 @@ def _xla_execute(fn, args, nprocs):
 @pytest.fixture()
 def xmp_executor():
     yield _xla_execute
+
+
+@pytest.fixture()
+def mock_gpu_is_not_available():
+    from unittest.mock import patch
+
+    with patch("torch.cuda") as mock_cuda:
+        mock_cuda.is_available.return_value = False
+        yield mock_cuda

tests/ignite/distributed/comp_models/test_native.py

@@ -32,6 +32,20 @@ def test__native_dist_model():
         assert "mpi" not in available_backends
 
 
+@pytest.mark.distributed
+@pytest.mark.skipif(not dist.is_nccl_available(), reason="Skip if nccl not available")
+def test__native_nccl_but_no_gpu(mock_gpu_is_not_available):
+
+    env_backup = os.environ
+
+    with pytest.raises(RuntimeError, match=r"Nccl backend is required but no cuda capable devices"):
+        _NativeDistModel(backend="nccl")
+
+    # environ could be corrupted by _NativeDistModel
+    os.environ.clear()
+    os.environ.update(env_backup)
+
+
 @pytest.mark.distributed
 @pytest.mark.skipif("WORLD_SIZE" in os.environ, reason="Skip if launched as multiproc")
 def test__native_dist_model_create_from_backend_bad_config():