Hi,

On Thu, Feb 21, 2008 at 12:19:55AM +0100, Johan Hoeke wrote:
> LS,
> 
> Running a 2 node cluster, heartbeat-2.1.3-3 centos rpms, RH AS 4.6
> 
> While testing a "maintenance scenario" for the cluster I set all
> resources to is_managed is false,
> 
> Feb 20 21:09:41 sierpinski pengine: [15725]: notice: native_print:
> R_BB10PRD_DB     (heartbeat::ocf:oracle):        Started
> sierpinski.uvt.nl (unmanaged)
> 
> 
> and proceeded to shut oracle by hand, oracle being one of the resources.
> 
> Feb 20 21:12:03 sierpinski oracle[23120]: [23145]: INFO: Oracle instance
> BB10PRD is down
> 
> 
> Within minutes, the node was stonithed. The log shows that this was
> right after the monitor operation for the oracle resource came back with
> return code 7:
> 
> Feb 20 21:12:03 sierpinski crmd: [4584]: info: process_lrm_event: LRM
> operation R_BB10PRD_DB_monitor_120000 (call=31, rc=7) complete
> 
> Feb 20 21:12:03 mandelbrot stonithd: [4580]: info:
> stonith_operate_locally::2375: sending fencing op (RESET) for
> sierpinski.uvt.nl to device external (rsc_id=R_ilo_sierpinski:0, pid=5414)
> Feb 20 21:12:03 mandelbrot stonithd: [4580]: info: Node
> mandelbrot.uvt.nl try to help node sierpinski.uvt.nl to fence node
> sierpinski.uvt.nl.
> 
> Conclusion: the monitor operation was still running even though the
> resource was unmanaged, and it forced a fencing action.

Oops. So there's an on_fail=fence for this monitor operation. Is
that necessary?

> I then made a script which in addition to changing the resources to
> is_managed = false also set the monitor operations to disabled=true.
> This worked, now I am able to shutdown oracle by hand without a fencing
> action starting up.
> 
> Questions:
> 
> It this expected behavior? Should monitor operations keep running even
> though the resources are set to is_managed=false?

Yes. There was some discussion about it and the majority of
votes went this way, i.e. that monitoring should continue even
for the unmanaged resources.

> Is explicitly setting
> the monitor operations to disable=true the "right way" to prevent
> unwanted fencing actions during cluster maintenance?

I'd say yes. But note that I was also in favour of having
monitoring disabled by default.

Thanks,

Dejan

> tia,
> Johan
> 
> (happy to post hb_reports if requested)
> 



> _______________________________________________
> 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