[Pacemaker] STONITH without mandatory success possible?

Vladislav Bogdanov bubble at hoster-ok.com
Thu Jun 20 09:02:18 CEST 2013


20.06.2013 09:00, Andrew Beekhof wrote:
> 
> On 20/06/2013, at 2:52 PM, Vladislav Bogdanov <bubble at hoster-ok.com> wrote:
> 
>> 20.06.2013 00:36, Andrew Beekhof wrote:
>>>
>>> On 20/06/2013, at 6:33 AM, Doug Clow <doug.clow at dashbox.com> wrote:
>>>
>>>> Hello All,
>>>>
>>>> I have some 2-node active-passive clusters that occasionally lose
>>>> Corosync connectivity. The connectivity is fixed with a reboot. They
>>>> don't have shared storage so stonith doesn't have to happen for another
>>>> node to take control of the resource. Also they are VMs so I can't use a
>>>> standard stonith method.
>>>
>>> fence_xvm works nicely for this most virt clusters, there was a vmware one floating around at one point
>>
>> Unfortunately matahari and libvirt-qmf (part of a picture, which
>> connects hypervisor to fenced) are removed in 6.4.
> 
> I didnt think matahari had anything to do with fence_xvm

libvirt-qmf depends on that - it connects to a QMF server using matahari
framework.

# rpm -e --test matahari-lib matahari-agent-lib
error: Failed dependencies:
        libmcommon.so.1.0.0()(64bit) is needed by (installed)
libvirt-qmf-0.3.0-10.el6.x86_64
        libmcommon_qmf.so.1.0.0()(64bit) is needed by (installed)
libvirt-qmf-0.3.0-10.el6.x86_64

There where libvirt-qpid, but it then was replaced with libvirt-qmf, and
fence_virtd was modified to use QMF server as well instead of plain qpidd.

So, one needs to build either libvirt-qpid or matahari+libvirt_qmf for a
hypervisor side, and have have either plain or qmf-patched version of
fence-virtd-libvirt-qpid, and to have either plain qpid of QMF
(matahari) server(s).

I just rebuilt matahari and libvirt-qmf from koji to use fence_xvm.

Current Redhat docs say that one needs to use fence_libvirt and have all
VMs pinned to a specific hypervisor host = no migration.

> 
>>
>>>
>>>> I could use the vmware module, but then if the VM host was down,
>>>> no node would take control. Is there a way to make stonith success not
>>>> mandatory?
>>>
>>> You can turn it off...
>>>
>>>> That way the other node would always take control and then it would 'try' to reboot the other machine if possible.
>>>>
>>>> On a related issue, CentOS 6.4 doesn't come with the null or ssh
>>>> devices. I've tried compiling from source but the resulting stonith
>>>> modules are not in the Red Hat type of format. Is there a way to get the
>>>> null or ssh device into CentOS?
>>>
>>> The RH ones are in the fence-agents package.
>>>
>>>
>>> _______________________________________________
>>> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>> Project Home: http://www.clusterlabs.org
>>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>> Bugs: http://bugs.clusterlabs.org
>>>
>>
>>
>> _______________________________________________
>> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>
>> Project Home: http://www.clusterlabs.org
>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>> Bugs: http://bugs.clusterlabs.org
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
> 




More information about the Pacemaker mailing list