NVIDIA / NVIDIA/nccl

NCCL data race

Open
#1,663 7 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

question triaged
Dominant language
C++
Stars
5.1k
Forks
1.4k
Avg merge
2h 3m
Merged PRs (30d)
2

Description

When trying to run Sanitizer for NCCL porting code, facing data race between threads realted proxy, following code call path.

Thread-A
proxy.cc: ncclProxyProgress →
proxy.cc: progressOps →
net.cc: sendProxyProgress
sub->sendMhandle = resources->mhandles[args->protocol];

Thread-B
proxy.cc: ncclProxyService →
proxy.cc: proxyProgressAsync →
net.cc: sendProxyConnect
net_ib.cc: ncclIbRegMr →
net_ib.cc: ncclIbRegMrDmaBuf
mhandle = (void) mhandleWrapper;

I would like to know that, is NCCL thread safe? If so, what will be the cause for the data race? If not, is there any way to avoid it?

Best Regards.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the reported call paths in proxy.cc, net.cc, and net_ib.cc, especially ncclProxyProgress, progressOps, sendProxyProgress, ncclProxyService, proxyProgressAsync, sendProxyConnect, ncclIbRegMr, and ncclIbRegMrDmaBuf. Reproduce the NCCL porting-code run under Sanitizer and inspect the accesses to resources->mhandles and mhandleWrapper. Done means identifying whether the report is a real race and documenting or resolving the supported thread-safety behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
distributed-systems, networking
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.