Tobias Appel wrote:
> On Fri, 2009-02-06 at 14:52 +0100, Dominik Klein wrote:
>> Just guessing, but your cluster does not know about those "external"
>> addresses, does it?
>>
> Well I have an internal ip added to /etc/hosts and a shortcut for the
> hostname, like nagios1 192.168.0.1 and then there is the full domain
> name like nagios1.hq.xxx.xxx 134.xxx.xxx.xxx and so on. 
> 
> Usually the cluster communication uses eth1 and the private IP adresses
> only! so when I pull that cable, network connectivity is still there but
> no communication between the cluster nodes except if you use the
> external ip adress. 
> 
> So I gave stonith_ssh as hostlist the full domain name.

So the node name is not in the hostlist. Therefore the cluster cannot
find a stonith device capable of managing that node.

Regards
Dominik

>> Sounds like you're only using one connection between the nodes for
>> cluster communication, pull that, see split-brain and want the cluster
>> to use a connection it does not know about.
> 
> true the cluster doesn't really know about the other connection, but as
> long as it uses the fulldomain name it should be able to connect to the
> other node using the external ip - at least when I do this manually it
> works.
> 
> In the logfiles it looks like it is trying to shoot each other, but like
> I said nothing is happening, but at the same time only one node had
> active resources - before implementing stonith_ssh both nodes would have
> run all resources at the same time.
> 
> And yeah atd is running (did a ps -aux | grep atd).
> 
> 
>> But even if you configured the cluster to also use the "external" path,
>> you'd have to unplug both to force them into a splitbrain scenario and
>> ssh stonith wouldnt help again (no useable, known connection left).
>>
>> Thats exactly why ssh stonith is bad.
>>
>> The only way stonith ssh "works" is for cluster software failure (like
>> pkill -9 heartbeat).
>>
> Ok that makes kinda sense, but still it should work in a 2-node setup,
> and it shouldn't be waiting for more members to vote?
> 
>> Regards
>> Dominik
>>
>> Tobias Appel wrote:
>>> Hi,
>>>
>>> I know I should use it only for development purposes and that's what I
>>> am doing right now.
>>>
>>> I have configured my cluster and a resource stonith_ssh. I configured
>>> stonith_ssh to use the external IP address to connect to the other
>>> server. When I ran: 'stonith -t ssh -p "nagios1.hq.xxx" -T reset
>>> nagios1.hq.xxx' it worked. The other node rebooted immediately. 
>>>
>>> But when I pulled the cross-over cable the results were weird. Each node
>>> had the same entry in the logfile which looked like this:
>>>
>>> Feb  6 12:57:59 nagios2 tengine: [27786]: info: te_fence_node: Executing
>>> reboot fencing operation (63) on nagios1 (timeout=30000)
>>> Feb  6 12:57:59 nagios2 stonithd: [27368]: info: client tengine [pid:
>>> 27786] want a STONITH operation RESET to node nagios1.
>>> Feb  6 12:57:59 nagios2 stonithd: [27368]: info: Broadcasting the
>>> message succeeded: require others to stonith node nagios1.
>>>
>>> Then later it said:
>>> Feb  6 12:58:29 nagios2 stonithd: [27368]: ERROR: Failed to STONITH the
>>> node nagios1: optype=RESET, op_result=TIMEOUT
>>>
>>> I'm not sure why it did not reboot the other machine. Does 'require
>>> others' mean that it won't work in a 2-node setup? 
>>>
>>> The really weird part is that even though no node was rebooted, only one
>>> node continued to run the resources, whereas the other was just doing
>>> nothing at all. When I reconnected the cross-over cable both nodes went
>>> into 'standby' state for a couple of seconds but everything was working
>>> fine.
>>>
>>> Regards,
>>> Tobi
>>>
>>> _______________________________________________
>>> 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
> 
> _______________________________________________
> 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