Event Model Design

Bryan Kearney bkearney at redhat.com
Tue May 18 19:45:42 UTC 2010


On 05/18/2010 03:39 PM, James Bowes wrote:
> On Tue, May 18, 2010 at 03:23:48PM -0400, Justin Harris wrote:
>>
>> ----- "Devan Goodwin"<dgoodwin at rm-rf.ca>  wrote:
>>
>>> Just sketching out some rough notes on the event model after
>>> conversations with Dmitri:
>>>
>>> https://fedorahosted.org/candlepin/wiki/EventModel
>>>
>>> This will be changing a bit but that's the general idea thus far.
>>> Lemme know what you think. Going to start prototyping and see what
>>> happens.
>>>
>>> Thanks,
>>>
>>> Devan
>>
>> We touched on this earlier, but how should we handle a failure/exception by a particular handler?
>>
>> We could either:
>>   - Keep going and silently ignore it
>>   - Treat each event as a transaction and roll everything back (this seems hard)
>>   - Maybe treat the default db persistence as a special case and only fire other events if this succeeds?
>>   - probably other ways
>>
>
> We're actually talking about event publication rather than the event
> itself (the event being, for example, creating a new consumer), right?
> So regardless of what happens during publication, the
> event has occured. In that case, I'd say we run all the publication
> actions on their own, and retry any that fail a configurable number of
> times.
+1 for Idempotency. +10 if you can actually pronounce it.

-- bk




More information about the candlepin mailing list