>>>>> "TM" == Trond Myklebust <trondmy@xxxxxxxxxxxxxxx> writes:
TM> Commit deaa5c96c2f7 ("SUNRPC: Address Kerberos performance/behavior
TM> regression") was supposed to be marked for stable as a fix.
I wonder, though; is that likely to be the root of the problem I'm
seeing? The commit description talks about this as a performance
regression, but I'm seeing a complete loss of NFS functionality.
Sadly I still don't have a reproducer, so outside of just deploying the
patch and hoping, I have no way to actually test this. So far I've been
running things like:
stress-ng --all 0 --class 'filesystem' -t 10m --times
in an NFS4-krb5p mounted directory without being able to reproduce the
problem. That drives the load up close to 200 but everything seems to make
progress. So it must be some specific sequence that causes it; I just
don't know which.
I did get this to show up in the kernel log, though, when I typed "df"
while running that stress-ng command was running:
[94547.656419] NFS: server nas00 error: fileid changed
fsid 0:57: expected fileid 0x6bc27eb, got 0x9996c6
I've never seen that one before, but it doesn't seem to hurt anything.
(This is still with 4.20.3.)
- J<
