[DRBD-user] Avoid blocking IO when node fails

Digimer lists at alteeve.ca
Thu Feb 13 19:51:02 CET 2014

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On 13/02/14 05:09 AM, Roberto Munoz Gomez wrote:
>
>
>
> -----Digimer <lists at alteeve.ca> escribió: -----
>
>> Para: Roberto Munoz Gomez/ES/DIA/DIAGROUP at DIAGROUP,
>> drbd-user at lists.linbit.com
>> De: Digimer <lists at alteeve.ca>
>> Fecha: 12/02/2014 20:47
>> Asunto: Re: [DRBD-user] Avoid blocking IO when node fails
>>
>> On 12/02/14 11:34 AM, Roberto Munoz Gomez wrote:
>>> Hi,
>>>
>>> I have configured an Active/Active cluster with Dual Primary DRBD.
>> All works fine and fast.
>>>
>>> But when a node fails, the DRBD status goes to WFConnection and all
>> IO is blocked on the other node. In this scenario I need the survivor
>> to continue using the DRBD partition. It would be great if when the
>> other node comes back, the synchronization begins automatically and
>> both become Primary automatically.
>>>
>>> I have search the doc and google, but have not found any of this.
>> Only the ko-count option, but does not seem to make a difference.
>>>
>>> Any ideas?
>>>
>>> Thanks in advance.
>>>
>>> Regards.
>>
>> You need fencing to safely proceed after a node loss. Are you using
>> cman
>> or pacemaker? If so, you can setup fencing in those (pacemaker calls
>> it
>> stonith, same thing) and then hook DRBD into it.
>>
>
>
> I'm using pacemaker and corosync. The fencing device is working correctly and is configured with:
>
> handlers {
> 	      fence-peer              "/usr/lib/drbd/crm-fence-peer.sh";
> 	      after-resync-target     "/usr/lib/drbd/crm-unfence-peer.sh";
> 	}
>
>
> The problem is that while the other node is being fenced, the active one cannot access the DRBD and all the IO calls are blocked until the fenced node comes alive again. And that is what I want to avoid. I want the survival node to continue working.

Do you have stonith configured and working in pacemaker? If not, this 
config won't work. The fence handlers simply pass the fence request to 
stonith and won't return success until/unless pacemaker says the fence 
succeeded.

-- 
Digimer
Papers and Projects: https://alteeve.ca/w/
What if the cure for cancer is trapped in the mind of a person without 
access to education?



More information about the drbd-user mailing list