Output to stderr in distributed tests. (pytorch#42139)

Summary:
Pull Request resolved: pytorch#42139

A bunch of tests were failing with buck since we would output to
stdout and buck would fail parsing stdout in some cases.

Moving these print statements to stderr fixes this issue.
ghstack-source-id: 108606579

Test Plan: Run the offending unit tests.

Reviewed By: mrshenli

Differential Revision: D22779135

fbshipit-source-id: 789af3b16a03b68a6cb12377ed852e5b5091bbad

Author: pritamdamania

test/distributed/test_c10d.py

@@ -40,7 +40,7 @@
 load_tests = load_tests
 
 if not c10d.is_available():
-    print('c10d not available, skipping tests')
+    print('c10d not available, skipping tests', file=sys.stderr)
     sys.exit(0)
 
 

test/distributed/test_c10d_spawn.py

@@ -20,12 +20,12 @@
 load_tests = load_tests
 
 if not c10d.is_available():
-    print('c10d not available, skipping tests')
+    print('c10d not available, skipping tests', file=sys.stderr)
     sys.exit(0)
 
 
 if NO_MULTIPROCESSING_SPAWN:
-    print('spawn not available, skipping tests')
+    print('spawn not available, skipping tests', file=sys.stderr)
     sys.exit(0)
 
 

test/distributed/test_distributed.py

@@ -116,7 +116,7 @@ def get_timeout(test_id):
 
 
 if not dist.is_available():
-    print("Distributed not available, skipping tests")
+    print("Distributed not available, skipping tests", file=sys.stderr)
     sys.exit(0)
 
 

test/distributed/test_nccl.py

@@ -1,5 +1,6 @@
 import unittest
 
+import sys
 import torch
 import torch.cuda.nccl as nccl
 import torch.cuda
@@ -18,7 +19,6 @@
 
 nGPUs = torch.cuda.device_count()
 if not TEST_CUDA:
-    import sys
     print('CUDA not available, skipping tests', file=sys.stderr)
     TestCase = object  # noqa: F811
 

test/test_cuda.py

@@ -37,7 +37,7 @@
 TEST_MULTIGPU = TEST_CUDA and torch.cuda.device_count() >= 2
 
 if not TEST_CUDA:
-    print('CUDA not available, skipping tests')
+    print('CUDA not available, skipping tests', file=sys.stderr)
     TestCase = object  # noqa: F811
 
 TEST_MAGMA = TEST_CUDA

test/test_cuda_primary_ctx.py

@@ -1,5 +1,6 @@
 import torch
 from torch.testing._internal.common_utils import TestCase, run_tests, skipIfRocm
+import sys
 import unittest
 
 # NOTE: this needs to be run in a brand new process
@@ -12,7 +13,7 @@
 TEST_MULTIGPU = TEST_CUDA and torch.cuda.device_count() >= 2
 
 if not TEST_CUDA:
-    print('CUDA not available, skipping tests')
+    print('CUDA not available, skipping tests', file=sys.stderr)
     TestCase = object  # noqa: F811
 
 

torch/testing/_internal/dist_utils.py

@@ -3,14 +3,15 @@
 import time
 from functools import partial, wraps
 import re
+import sys
 
 import torch.distributed as dist
 import torch.distributed.rpc as rpc
 from torch.distributed.rpc import _rref_context_get_debug_info
 
 
 if not dist.is_available():
-    print("c10d not available, skipping tests")
+    print("c10d not available, skipping tests", file=sys.stderr)
     sys.exit(0)