On Thu, Dec 3, 2009 at 8:53 AM, Michael Schwartzkopff <[email protected]> wrote:
> Hi,
>
> I was looking into the SystemHealth feature a little bit deeper and I noticed
> perhaps a little design error:
>
> Given you have a HealthAgent that monitors the CPU load on one node and the
> Health-Strategy set to "migrate on red". Now you clone that Agent to measure
> the CPU load on all  nodes.
>
> Now the Load on one system goes up, the #health-cpu attribute goes "red" and
> all resources move away from the node. Of course the instance of the
> SystemHealth clone that measures the load also moves away from that node. Thus
> it cannot measure the cpu load of the node any more and the attribute stays on
> "red" even if the load drop below the limits.
>
> The admin must remove attribute manually to allow the clone instance (and all
> other resoruces) to be started on that node again. Not a very nice situation.
>
> Perhaps it would be better the make all Health-RAs to ignore the Health-
> Strategy.

Agreed.
Could you create a bugzilla for this so I don't forget?

I had been thinking about a new resource class (ie. instead of ocf)
but I think that will create more confusion than its worth (since the
same RAs will also show up as regular resources).  Instead I think we
need a new meta option that basically tells the PE that the resource
should always be active as long as the node is up (something we
probably want for stonith resources too).
_______________________________________________
Openais mailing list
[email protected]
https://lists.linux-foundation.org/mailman/listinfo/openais

Reply via email to