Data race in peer about nuraft

[greensky00]

From what you provided, it seems there is a race between srv_to_join_.reset() and peer::lock_. However, that is not possible as peer::handle_rpc_result has myself not to make the reference counter of the shared pointer become 0. I haven't seen TSAN alert during the normal server add/removal process.

• Can you please provide the exact commit hash of NuRaft that you used?
• Also, do you get this TSAN alert every time you run the above code? If so, it will be great to share the logs generated by NuRaft to see what happened.

[as51340]

Thanks for the reply!

We are using NuRaft v2.1.0
Yes, TSAN alerts on every run.

Can you instruct me on how to get logs?

[greensky00]

I tried the same code with the same NuRaft version, still didn't get TSAN alert.

Could you share your build environment? Such as OS version, compiler version, etc. Also please share Raft parameters (heartbeat, election lower/upper timeout, ...).