On Thu, 2017-11-02 at 08:57 +0000, Marian Rainer-Harbach wrote:
I'm sorry it took me so long to get back to you! I get what you're
saying and it seems like a possible explanation for some scenarios.
However, I'm still not sure it applies to my case, where 389 DS
behaves completely normal for a few minutes and then an instantaneous
jump to full CPU usage occurs.
I created three pstack traces:
was taken while the load test was
already running and 389 DS still behaved normally.
is one minute later. The same load
test is still running, but now 389 DS uses all CPUs and is very slow.
is after stopping the load test, 389
DS is now idle.
It would be great if you could take a look at those traces!
Thread wise, I can see that the time in pbkdf2 is likely your issue.
You are not consuming all workers though, as you can see many threads
are in pthread_cond_wait/connection_threadmain.
However, I still think the initial assessment was correct. Have you
tried out the latest version with the pbkdf2 timing changes applied?
Hope that helps!
Red Hat, Australia/Brisbane