ons 2013-07-24 klockan 23:35 +0200 skrev squadra:

Maybe found a workaround on the NFS server side, a option for the mountd service

     -S      Tell mountd to suspend/resume execution of the nfsd threads when-

             ever the exports list is being reloaded.  This avoids intermit-

             tent access errors for clients that do NFS RPCs while the exports

             are being reloaded, but introduces a delay in RPC response while

             the reload is in progress.  If mountd crashes while an exports

             load is in progress, mountd must be restarted to get the nfsd

             threads running again, if this option is used.

so far, i was able to reload the exports list twice, without any random suspended vm. lets see if this is a real solution or if i just had luck two times.

It would seem as if we are on the same boat:) Actually I hadn´t thought about it before, but you´re right; issuing a "service mountd reload" does pause a large number of VM´s, frickin annoying really. I mean, the NFS server doesn´t care what or who it´s serving, you could be creating a new export for a completely different system, and not even have oVirt in mind before customers start to call, wondering why their VM´s have stopped responding!?

I actually tried that "-S" but it didn´t work for me at all, and looking at the man-page for mountd, there´s no mention of it either, even though we are presumably running the same version:
# uname -r
9.1-RELEASE

Or are you perhaps tracking "-STABLE", and there´s a minor difference there?

but i am still interested in parameters which make the vdsm more tolerant to short interruptions. instant suspend of a vm after such a short "outage" is not very nice.

+1! 

/Karli