From d35a600b74dc8cb4ddcc39e07aa309325affaab9 Mon Sep 17 00:00:00 2001
From: Shuqiang Zhang <sqzhang@meta.com>
Date: Tue, 5 Nov 2024 15:16:35 -0800
Subject: [PATCH] [pgnccl] skip restart test fro rocm (#139809)

Summary:
PG restart test is flaky in rocm: https://github.com/pytorch/pytorch/pull/139809, skip the AMD/ROCM test for now
Test Plan:
CI

Tags:

Pull Request resolved: https://github.com/pytorch/pytorch/pull/139809
Approved by: https://github.com/kwen2501
---
 test/distributed/test_c10d_nccl.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/test/distributed/test_c10d_nccl.py b/test/distributed/test_c10d_nccl.py
index 0dd46749b7c..f5f971c6ae7 100644
--- a/test/distributed/test_c10d_nccl.py
+++ b/test/distributed/test_c10d_nccl.py
@@ -348,6 +348,7 @@ class ProcessGroupNCCLGroupTest(MultiProcessTestCase):
             dist.all_reduce(t)
 
     @requires_nccl()
+    @skip_if_rocm_multiprocess
     @skip_but_pass_in_sandcastle_if(not TEST_MULTIGPU, "NCCL test requires 2+ GPUs")
     def test_restart_pg(self):
         # Note: restart test passes steadily only for blocking mode for now.