Friday, 31 October 2014

Virtlockd on Ubuntu Server 14.10

Virtlockd shouldwork out of the box on 14.10 but it doesn't!

To implement locking on SAN volumes in a 3 node QEMU/KVM cluster I wanted to implement virtlockd.  I first set up a glusterfs cluster for the three nodes to share a common filesystem to store the lock files and mounted this at /var/lib/libvirt/lockd on each node.

I turned on locking in /etc/libvirt/qemu.conf by uncommenting the line:
lock_manager = "lockd"

Set auto_disk_leases = 1 in /etc/libvirt/qemu-lockd.conf and uncommented all of the default lockspace_dir entries.

I restarted libvirt with sudo service libvirt-bin restart and attempted to start a machine.

I was met with the error:
Failed to connect socket to '/var/run/libvirt/virtlockd-sock': No such file or directory

After searching in various config files I found that virtlockd needs to start as a daemon and there was no auto way for this to happen so I wrote an upstart config...

description "virtlockd Daemon"
author "Guy Thouret <guy@thouret.uk>"

start on stopped rc RUNLEVEL=[2345]
stop on starting rc RUNLEVEL=[016]

expect daemon
respawn

exec /usr/sbin/virtlockd -d

...saved it as /etc/init/virtlockd.conf and I was then able to start virtlockd with service virtlockd start

I confirmed the lockfile had indeed been created:
guy@cyclopes:/gluster/vmlocks$ sudo ls -l /var/lib/libvirt/lockd/*
-rw-r--r-- 1 root root     0 Oct 30 14:02 /var/lib/libvirt/lockd/tsting

/var/lib/libvirt/lockd/files:
total 0

/var/lib/libvirt/lockd/lvmvolumes:
total 0

/var/lib/libvirt/lockd/scsivolumes:
total 0
-rw------- 1 root root 0 Oct 30 16:33 3600c0ff0001587169836e15201000000

Attempting to start the running VM on another node gave the error:
Error starting domain: resource busy Lockspace resource '3600c0ff000158716ec34b05201000000' is locked

Virtlockd has saved the day and prevented the VM from potentially being trashed.

No comments:

Post a Comment