I've had a similar problem this morning - however, typing 'sync' didn't
help - nor did rebooting the server - after a reboot the nfsd's worked
for a short while, then they 'stalled' again.
The server has two XFS volumes (/disk1 and /disk2), running 'find
/disk2' on the server also 'hung' ..., so I brought the machine up in
single user mode and ran xfs_repair - but it reported no problems -
however when I then brought the machine up in multi user mode, NFS
access was OK - I don't know if running xfs_repair helped, or if it was
just a coincidence that whatever NFS access that had been causing the
stall 'went away'...
If this still sounds like the problem that Keith has described, then I
would like to know if the fix can be back ported to 2.4.18/XFS1.1 - I
don't want to use a CVS version unless I really have to ...
Thanks
James Pearson
Keith Owens wrote:
>
> On Fri, 05 Jul 2002 10:55:45 +0100,
> James Pearson <james-p@xxxxxxxxxxxxxxxxxx> wrote:
> >Do you happen to know if the fix can be back ported to 2.4.18/XFS 1.1 ?
> >Where in the CVS code would I find the fix?
>
> I was hoping you would not ask that ;). The CVS tree has five months
> of changes since 2.4.18, both to XFS and to go from 2.4.18 to
> 2.4.19-rc1. I don't know which fix corrected this problem. Steve Lord
> might know but he is away until July 8, USA time.
> On Thu, 04 Jul 2002 17:16:06 +0100,
> James Pearson <james-p@xxxxxxxxxxxxxxxxxx> wrote:
> >We have noticed a problem with a couple of our NFS servers (running
> >RedHat 7.2 with a stock 2.4.18 kernel with XFS v1.1) whereby NFS access
> >slows to a crawl or stalls.
> >
> >The exported filesystem(s) are XFS with 8 nfsd's running - when we have
> >the problem the load average is about 8 - but CPU usage, disk access and
> >network traffic are minimal.
> >
> >I found, by accident, that running the command 'sync' appears to 'fix'
> >the situation...
> >
> >I'm not sure if this is an XFS or NFS related problem (hence posting to
> >both lists).
>
> XFS. 2.4.18 would sometimes get into a situation where two XFS
> operations were waiting on locks (not deadlocked) and nothing was
> moving. Performing some other disk activity such as sync would get
> things moving again.
>
> AFAICT this is fixed in the XFS CVS tree, against 2.4.19-rc1.
|