NFS-shared file-system is locking up

Our NFS-shared file-system is locking up.

Please feel free to ask any questions you feel relevant. 🙂

At the time, there are a lot of processes in "disk sleep" state, and the load averages on our machines sky-rocket. The machines are responsive on SSH, but our the majority of our websites (apache+mod_php) just hang, as does our email system (exim+dovecot). Any websites which don't require write access to the file-system continue to operate.

The load averages continue to rise until some kind of time-out is reached, but for at least 10-15 minutes. I've seen load averages over 800, yet the machines are still responsive for actions which don't require writing to the shared file-system.

I've been investigating a variety of options, which have all turned out to be red-herrings: nagios, proftpd, bind, cron tasks.

I'm seeing these messages in the file server's system log:

Jul 30 09:37:17 fs0 kernel: [1810036.560046] statd: server localhost not responding, timed out
Jul 30 09:37:17 fs0 kernel: [1810036.560053] nsm_mon_unmon: rpc failed, status=-5
Jul 30 09:37:17 fs0 kernel: [1810036.560064] lockd: cannot monitor node2
Jul 30 09:38:22 fs0 kernel: [1810101.384027] statd: server localhost not responding, timed out
Jul 30 09:38:22 fs0 kernel: [1810101.384033] nsm_mon_unmon: rpc failed, status=-5
Jul 30 09:38:22 fs0 kernel: [1810101.384044] lockd: cannot monitor node0

Software involved:

VMWare, Debian lenny (64bit), ancient Red Hat (32 bit) (version 7 I believe), Debian etch (32bit)

NFS, apache2+mod_php, exim, dovecot, bind, amanda, proftpd, nagios, cacti, drbd, heartbeat, keepalived, LVS, cron, ssmtp, NIS, svn, puppet, memcache, mysql, postgres

Joomla!, Magento, Typo3, Midgard, Symfony, custom php apps