(In reply to Yaniv Kaul from comment #16)
> What's the next step here?
(Copying from https://bugzilla.redhat.com/show_bug.cgi?id=1721355#c9)
<snip> we need to add better eager-lock debugging infra to AFR. Since https://review.gluster.org/19503, AFR maintains multiple queues for fops for performance gains and it is a bit difficult as of now to gain insight into when eager locks are acquired and released when there is a lot of I/O being pumped. </snip>
Also, we need to write some python gdb helper scripts (my question on stack overflow (comment #13) has since received an answer) to isolate threads of interest in a multi-hundred thread process. I'm currently working on the lock less heal info bug. I can pick this up once that is complete.