[ClusterLabs] NFS cluster fails over ok but client mountpoint hangs and cluster VMs go crazy on host

Dennis Jacobfeuerborn dennisml at conversis.de
Sun Jun 14 21:12:10 EDT 2015

I've created a straightforward NFS cluster (basicall two KVM guests on
my laptop both running centos 7.1) and the failover works fine on the
server side but when I initiate a failover then an "ls <mountpoint>"
hangs for quite a while after the server-side is back online and more
suspiciously the cpu usage of the cluster guests and their associated
vhost processes goes through the roof. Curiously the cpu usage is not
visible inside the guests. Eventually (sometimes after just 20 seconds,
sometimes after a minute) things calm back down again and I get the
directory listing.
This also happens when I wait until the failover is completed (without
accessing the mountpoint during the failover) and only then access the
mountpoint again.
Does anyone have an idea what could cause behavior like this? I mounted
the NFS on both the host system and also in a separate guest and in both
cases I see similar behavior. The host is a Fedora 21 system.

This is what the exportfs resource on the server side looks like:
 Resource: media-live-export (class=ocf provider=heartbeat type=exportfs)
  Attributes: directory=/srv/exports/media-live fsid=2
unlock_on_stop=true wait_for_leasetime_on_stop=true
  Operations: start interval=0s timeout=40
              stop interval=0s timeout=120
              monitor interval=10 timeout=20

This is how I mount the NFS on the client:
mount -t nfs4 -o
nfsvers=4,noatime,hard,intr,rw,timeo=250,retrans=30,sec=sys /mnt/tmp


More information about the Users mailing list