[Libvirt-cim] [PATCH 0 of 2] [RFC] ComputerSystemModifiedIndication

[Jay Gagnon]

Heidi Eckhart wrote:
> Dan Smith wrote:
>> JG> Is there a mechanism for indications reporting errors, beyond
>> JG> CU_DEBUG'ing "help" and hoping to recover?
>>
>> I would expect that the answer will be that there isn't much that can
>> be done.  A CU_DEBUG() indicating the issue is certainly required, but
>> I'm not sure what else we have.
>>   
> CU_DEBUG() is the providers possibility to report an error. A
> mechanism to report a failure in the indication processing itself to
> the subscriber of the ComputerSystemModfiedIndication can not be done
> via the ComputerSystemModifiedIndication. This indication will only be
> thrown, if all went fine. So an indication in general is only
> reporting positive results - means, if the filter requirement is
> fulfilled, the indication is thrown. To monitor a failing indication
> processing an additional indication provider would be necessary. This
> one would then handle "please inform me, if something in the
> processing of the ComputerSystemModfiedIndication failed". So a client
> would need to subscribe 1) to the ComputerSystemModifiedIndication to
> get informed of the modification and 2) to the AlertIndication
> monitoring the indication processing to get informed of a failing
> processing.
Excellent explanation, thanks for that. :)  Fortunately it matches
pretty closely to my assumption for how things would work, which is
always nice.
>> I tend to think that indications are mostly for convenience and
>> optimization, to avoid pinging the providers unnecessarily, in which
>> case an error wouldn't be catastrophic.  Multiple clients monitoring a
>> single provider would use them to notice when the other makes a
>> change.
>>   
> I absolutely agree. For now I would say, its more important to have
> the ComputerSystemModifiedIndication. But at a later point in time -
> an indication provider who reports processing failures could be added
> as a feature.
>
Good, we're all on the same page here.  I'll worry about the failure
reporting later (I'll leave the CU_DEBUG calls in for our testing
convenience).

-- 

-Jay