prov/verbs: Fix data race vrb_open_ep function

[piotrchmiel]

Race Fix: Resolved a data race in the vrb_open_ep function of the verbs provider caused by concurrent modifications to the global variable vrb_ep_ops. This issue violated the FI_THREAD_SAFE threading model, leading to unpredictable behavior when creating endpoints from multiple threads.

Detailed race description: #10569

Signed-off-by: Piotr Chmiel [email protected]

[shefty]

Please change the commit message header to: prov/verbs: Fix race in ..., then describe the actual race in the message.

[piotrchmiel]

Please change the commit message header to: prov/verbs: Fix race in ..., then describe the actual race in the message.

Done. @shefty Could one of the maintainers kindly review this?

[shefty]

Thanks - The current code is incorrect, and this looks like the right fix.

[piotrchmiel]

@shefty Thanks for the review! I appreciate your feedback, and I’m glad the change looks correct now.
Could you clarify what the next steps are in the merging process? I noticed that the AWS CI check is failing with the message “This commit cannot be built.” However, when I click on the workflow details, the page doesn’t load, and I’m unsure if it's a permissions issue on my side.

[shefty]

bot:aws:retest

[shefty]

@piotrchmiel - This PR should eventually be merged. I restarted the AWS CI, but I don't believe the previous failure was related.

[shijin-aws]

we disabled verbs in aws ci, please ignore then

[shefty]

@shijin-aws - Is there a way for it to ignore or skip the testing, but report success? I believe other CI's do this.

[shijin-aws]

@shefty That is a good call, we will evaluate it.

[shijin-aws]

How does other CI handle it as scripts? is there an example?

[shefty]

See:

libfabric/contrib/intel/jenkins/Jenkinsfile

Line 323 in d2f7028

def skip() {