[Linux-cluster] Why token value > quorum_dev_poll ??

umesh susvirkar susvirkar.3616 at gmail.com
Mon May 17 06:40:17 UTC 2010


On Wed, May 12, 2010 at 10:16 PM, umesh susvirkar
<susvirkar.3616 at gmail.com>wrote:

> is this mandatory .
>
>
> i did not face any problem even if quorum_dev_poll value > totem
>
> but redhat always say quorum_dev_poll value should be  < totem
>
>  Can any one give me scenario were quorum_dev_poll value > totem will
> create a problem but not with quorum_dev_poll value < <totem
>
>
> they  give me this scenario.
>
>
> if qdiskd hang & at the same time my network go down
>
> What i found
> case 1: quorum_dev_poll =40 s totem=20s
>
> what i did i killed the qdisk process & at the same remove the network
> cable
>
> output was: at 20 s token was lost & both node fenced each other
>
> case 2: quorum_dev_poll= 20s totem=30s
>
> at 20s both lost contact with quorum device but no problem in cluster,but
> at 30s token lost as well as quorum was dissolve ( no fencing happen )
> i have to manually reboot the server
>
>
> what i got explanation  for case 1 is that  the quorum process is dead it
> vote still present with cman
> i agree with this  but quorum vote is really matter when your heartbeat
> packet is lost.
>
> Case 2: cluster is behaving as expected
>
> if you can see in both the case the actual decision is taken when the token
> value is time out,then were come dependency between
> 2 parameter.
>
>
>  You can check the log attach for your reference ( u can search for string
> umesh in node1 file )
>
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20100517/10d22cba/attachment.htm>


More information about the Linux-cluster mailing list