[ovirt-users] ovirt and gluster storage domain HA testing

Nir Soffer nsoffer at redhat.com
Wed Feb 17 02:13:44 EST 2016


On Wed, Feb 17, 2016 at 7:57 AM, Sahina Bose <sabose at redhat.com> wrote:
>
>
> On 02/17/2016 05:08 AM, Bill James wrote:
>>
>> I have a ovirt cluster with and system running ovirt-engine 3.6.2.6-1 on
>> centos7.2
>> and 3 hardware nodes running glusterfs 3.7.6-1 and centos7.2.
>> I created a gluster volume using gluster cli and then went to add a
>> storage domain.
>>
>> I created it with Path ovirt1-ks.test:/gv1 and all works fine, until
>> ovirt1 goes down.
>> Then ALL VMs pause till ovirt1 comes back up.
>> Do I have to list all nodes in the path for this to work?
>>
>> Path: ovirt1-ks.test:/gv1 ovirt2-ks.test:/gv1 ovirt3-ks.test:/gv1
>>
>> Or how do I prevent ovirt1 from being single point of failure?
>
>
> Which type of storage domain have you created - NFS or GlusterFS?
> With NFS, this could be a problem, as the nfs server running on ovirt1 can
> become single point of failure. To work around this, you will need to set up
> HA with CTDB or pacemaker/corosync depending on version of NFS you're using.
>
> If you're using glusterfs, did the hypervisor node go down as well when
> ovirt1 went down? Only during mounting of the storage domain (either on
> activate of hypervisor host/ reboot of host) is the server provided in path
> required to access volume (ovirt1). You can provide
> backup-volfile-servers=ovirt2-ks:ovirt3-ks in Mount options while creating
> the storage domain.

oVirt is automatically detecting the available bricks and generating the
backup-volfile-servers option when connecting to a gluster storage domain.

You can see if this worked in vdsm.log, the backup-volfile-servers option
should appear in the mount command.

Please attach vdsm.log showing the time you connected to the gluster
storage domain, and the time of the failure.

Attach also /var/log/sanlock.log - this is the best place to detect issues
accessing storage as it reads and writes to all storage domains frequently.

> Please also provide output of gluster volume info. (I'm assuming the bricks
> on the remaining servers were online)

Nir


More information about the Users mailing list