On Fri, Aug 7, 2009 at 11:57 AM, Alain.Moulle<[email protected]> wrote:
> Hi,
>
> ok but do you agree that in case of heartbeat network problem, there will
> be a "race to stonith" from all nodes in the cluster and so the risk that
> both nodes will be killed is not zero ?

not zero but incredibly small.

first of all, bot all nodes have the same amount of work to do.  one
node usually has a head start.
second, some switches only allow a single connection, perhaps that is
a feature in your case.

also, this only happens when you've set no-quorum-policy=ignore which
is really not recommended.

> That's why I thought that a ping towards an equipment out of the cluster
> should reduce the risk of split brain : suppose that each node pings its
> Eth switch
> (each node connected to a different Eth switch) , and suppose that there is
> a network problem on one side only, the node which has problem will
> not ping and will suicide itself, whereas the node which will ping the
> Eth switch will not suicide and will stonith the other one.
> Do you agree with this "theory" ?

Have you read about pingd?

> Thanks
> Alain
>> And how should we proceed to avoid split-brain cases in a two-nodes
>> > cluster  in case
>> > of problems on heartbeat network ?
>>
>>
>> make "network" "networks" (plural) to reduce the chance of getting into
>> a split-brain sitatuation and get and configure stonith devices to
>> protect your data in case it happens anyways.
>>
>> Regards
>> Dominik
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to