Open Stack

> I'd say events are *more* useful in that workflow, not less, as long as  
> they contain enough context. For example, the user creates a volume,  
> tries to attach it which fails for some config error, so the user  
> deletes it. With an event based model, the admin now has an error event  
> in their queue. If we used a db field then the error status is  
> potentially revived by the successful delete. 

+1

Nova currently emits a good set of events and errors and we've found it especially useful to debug / do postmortem analysis by collecting these notifications and being able to view the entire workflow. we've found quite a few occasions where the error popups presented in Horizon are not the real error but just the last/wrapped error.

there are various consumers that already collate these error notifications from Nova and i don't think it's much of a change if any to collect error notifications from Cinder. i don't think there's any change from Ceilometer POV -- just publish to error topic.

cheers,

gord