Frank Filz

So I'm confused about the different latencies here... What has been broken down so far as to where time is spent?

This really needs more investigation but I'm not sure where the resources are to do it.

Bump so the live issues are on the first page

I'm not sure where to go on this one. There is almost assuredly performance work that could be done but we don't have the resources. I will leave this open...

What operations are in the compound? Are there any errors in the log?

Where in thread2 did the release_slot call come from?

Hmm, so the cached_result is refcounted.... Maybe SEQUENCE has to actually do the res_compound4_extended swap while the lock is held, then the refcount would protect the response even if the...

If you can run Ganesha with valgrind/massif that can help determine where the memory is going. We have in the past had issues where any attempt to understand the memory...

What is your config? It looks like you may have a lot of mdcache entries with ACLs. It would also be worth looking into the massif tool which can capture...

Looks like you're missing nfs-ganesha debug information which makes the Ganesha part of the massif stack dumps less useful. Looks like huge amounts of Gluster memory allocations, but not quite...