Neil Horman

was it perhaps the KLT disablement of rwlocks in https://github.com/openssl/openssl/pull/24969 ?

I think the only critical fix for threading in that change was https://github.com/openssl/openssl/pull/24969/commits/5381bc8f0c155162a2e93fe6e974e3977fd132b8 , but that was for the KLT model, not the PUT model as you are requesting. All...

ACK, do you want to close this, or should we rename it to track down the current failures?

To clarify, is this running with the PUT thread model, or the new KLT thread model? Focusing on the first error: ``` ERROR: (bool) 'globserverret == true' failed @ /home/jenkinsbuild/.jenkins/workspace/OpenSSL-3.4_Monitor/test/helpers/quictestlib.c:572...

Focusing on the quic test So that tells us the thread is running properly, but getting a legitimate error when creating the quic connection. Will need to add some instrumentation...

what changed? Or is this just changing from run to run with the same build?

can you try build this branch: https://github.com/nhorman/openssl/tree/lock_trace and run the following: ``` make TESTS=test_threads V=1 OPENSSL_TRACE=LOCK test ``` capturing the output (warning, it will likely be lots of text) Lets...

hmm, me too. I'm not strictly suprised by the fact there was no output, because (and this is more suprising), I expected the hang to occur in either the rcu...

the balancing hierarchy was numa_node->package->cache->core The selection for an appropriate numa node is handled in the do_one_cpu function, restricting the cpu selection to the intersection of the package mask and...

Early estimates say about two minutes, but I need to convert the other builds to get a better average