Open Stack

Recently I've been focused on ensuring we don't drop notifications in
CM. But problems still exist downstream, after we've captured the raw
event.

>From the efforts going on with the Ceilometer sample pipeline, the new
dispatcher model and the upcoming trigger pipeline, the discussion
around retry semantics has being coming up a lot.

In other words "What happens when step 4 of a 10 step pipeline fails?"

As we get more into processing billing events, we really need to have a
solid understanding of how we prevent double-counting or dropping events.

I've started writing down some thoughts here:
https://wiki.openstack.org/wiki/DuplicateWorkCeilometer

It's a little scattered and I'd like some help tuning it.

Hopefully it'll help grease the skids for the Icehouse Summit talks.

Thanks!
-S

cc/ Josh, I think the State Management team can really help out here.