https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=249871
--- Comment #6 from Rick Macklem <rmack...@freebsd.org> ---
One additional thing you could do is:
# vmstat -z | fgrep mbuf_cluster
on the server when it is under heavy load,
to see if it is running out of mbuf clusters.
(Theoretically the NFS server should keep
working when mbuf clusters are exhausted,
but the result would be a Readdir reply
made up of a long list of regular mbufs.
That could impact things like TSO, if the
net interface on the server has that
enabled.)
Basically, other than possible mbuf exhaustion,
I can't think of any way heavy load would
affect the NFS server code (except slower
response).
Since the name cache doesn't seem to be the
culprit, that leaves all the caching that
goes on inside ZFS.
--> If the readdir contents is somehow reordered
by ZFS when the directory is under heavy
readdir load or the directory offset cookies
somehow change, that would explain the
problem.
Yet one more thing that you could try is having
client mounts done with "nfsv3,rdirplus".
You mentioned that NFSv3 worked ok.
NFSv3 + ridrplus works more closely to NFSv4
in the server, and whether or not that fails
might be useful information.
--
You are receiving this mail because:
You are the assignee for the bug.
_______________________________________________
freebsd-bugs@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-bugs
To unsubscribe, send any mail to "freebsd-bugs-unsubscr...@freebsd.org"
