Re: Processes stuck in D state..

[Keith Owens <kaos@sgi.com>]

On Wed, 6 Aug 2003 09:03:26 +0100 (BST), 
Gordon Henderson <gordon@drogon.net> wrote:
>
>I am running a few production servers with XFS now, but I'm a little
>concerned... (as I'm seeing some problems)
>
>Anyway, I'm using the -ac4 patches to 2.4.21 with the 1.2 release of the
>xfsprogs and with overnight cron "stuff", (eg. amanda)  I get a bunch of
>processes hanging in the "D" state. Eg. xfsdump. This doesn't inspire
>confidence. Right now I'm seeing:
>
> 1149 ?        D      2:20 xfsdump -F -J -l 1 - /dev/md4
> 9275 ?        D      2:02 du -k
>10103 ?        D      2:26 /usr/bin/find / ( -fstype NFS -o -fstype nfs -o
>21671 ?        D      1:28 /usr/bin/rsync -a --delete -H -x . /mounts/local0.yesterday/
>13976 ?        D      1:16 du -k
>14776 ?        D      1:59 /usr/bin/find / ( -fstype NFS -o -fstype nfs -o
>
>which isn't good. A few days ago I had a bunch of nfsd's stuck in D state
>too which required a reboot - which in a busy working environment isn't a
>good thing )-:

Do not blame XFS for this problem, it is almost certainly an NFS
problem.  When an NFS mount marked 'hard' stops responding, any process
that accesses the NFS mount will hang in D state.  If those processes
are holding mount or vfs locks, then other processes that require mount
activity will hang on those locks, also in D state.

Fix the NFS problems and the other hangs will go away.  If NFS hangs a
lot, try mounting with 'hard,intr' which lets you kill the hung tasks,
or even 'soft,intr' so they automatically time out on hung NFS mounts.
Note: a 'soft' NFS mount can result in I/O errors when the NFS server
is unreliable.