[Pacemaker] Three questions...

Andrew Beekhof beekhof at gmail.com
Thu Mar 19 08:09:26 UTC 2009


On Thu, Mar 19, 2009 at 07:23, Takenaka Kazuhiro
<takenaka.kazuhiro at oss.ntt.co.jp> wrote:
> Hello
>
>> It is completely dependent on your stonith architecture.
>> Some devices support a list of hosts which means you only need one
>> stonith resource.
>
> Even if there is a such device, the Pacemaker stonith mechanism
> provides insufficient support for it. Suppose: how should its plugin
> instance do when it loses control to one of its target nodes? If a
> monitor for the instance returns an error code, Pacemaker loses control
> via it to all of its target nodes at a time.

on-fail=ignore ?

> On the other hand, If a
> monitor for the instance returns a success code, Pacemaker cannot
> be aware of the trouble. In the later case, when the node that the
> instance cannot control must be shooted, Pacemaker will request
> the useless instance to shoot the node.
>
> Andrew Beekhof wrote:
>>
>> 2009/3/16 Romi Verma <romi3rdfeb at gmail.com>:
>>>>
>>>> i guess if DC fails then CRM will elect any other node for DC role and
>>>> then , this new DC  will try to fence the errant node , if it fails to
>>>> fence
>>>> then it will choose other node for this fencing task.
>>>
>>>  Andrew could you please confirm the above .
>>
>> [insert same rant as Lars - this is getting _really_ annoying]
>>
>>> one thing more related to stonith design. suppose we have 4 nodes cluster
>>> so
>>> do we need 3 stoniths on each node ( to make a node eligible to kill any
>>> of
>>> other 3 nodes ) that will lead to 3x4 stoniths or  do we need only  one
>>> stonith on  each node that will be total 4 stoniths .
>>
>> It is completely dependent on your stonith architecture.
>> Some devices support a list of hosts which means you only need one
>> stonith resource.
>>
>>> i experimented this and found that 3 stoniths on each node is required
>>> but i
>>> guess it should not be the c ase as CIB has the stonith info so even if
>>> stonith is not running on a node it shoul be able to  read the stonith
>>> related info from CIB.
>>>
>>> Thanks,
>>>>
>>>> Thanks & Regards,
>>>>
>>>>>
>>>>> Regards,
>>>>>
>>>>> _______________________________________________
>>>>> Pacemaker mailing list
>>>>> Pacemaker at oss.clusterlabs.org
>>>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Pacemaker mailing list
>>>>> Pacemaker at oss.clusterlabs.org
>>>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>>
>>>>
>>>> _______________________________________________
>>>> Pacemaker mailing list
>>>> Pacemaker at oss.clusterlabs.org
>>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>
>>>
>>> _______________________________________________
>>> Pacemaker mailing list
>>> Pacemaker at oss.clusterlabs.org
>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>
>>>
>>
>> _______________________________________________
>> Pacemaker mailing list
>> Pacemaker at oss.clusterlabs.org
>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>
>
> --
> Takenaka Kazuhiro <takenaka.kazuhiro at oss.ntt.co.jp>
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>




More information about the Pacemaker mailing list