Revert "[PGNCCL] Launch kernel on current stream & remove record_stream entirely (#148590)"

This reverts commit 17dbeb11db.

Reverted https://github.com/pytorch/pytorch/pull/148590 on behalf of https://github.com/janeyx99 due to PR break backward compat test ([comment](https://github.com/pytorch/pytorch/pull/148590#issuecomment-2708641172))
This commit is contained in:
PyTorch MergeBot
2025-03-09 03:01:55 +00:00
parent 17dbeb11db
commit 9cb25f0ea2
10 changed files with 362 additions and 408 deletions

View File

@@ -363,9 +363,6 @@ class TestDebugInfoWriter : public c10d::DebugInfoWriter {
};
TEST_F(ProcessGroupNCCLErrorsTest, testNCCLErrorsNoHeartbeat) {
// Note (kwen2501) 03/07/2025
// TODO: re-enable
GTEST_SKIP() << "Skipping test as the trace write seems unstable.";
int heartBeatIntervalInSec = 2;
std::string timeInterval = std::to_string(heartBeatIntervalInSec);
ASSERT_TRUE(setenv(c10d::TORCH_NCCL_BLOCKING_WAIT[0].c_str(), "0", 1) == 0);