Open Stack

On 22/10/16 10:38, Thomas Herve wrote:
> Hi all,
>
> One of my long time goal since I started contributing to OpenStack is
> to try to remove polling where I can. With Zaqar WebSocket support, we
> now have a transport available for users to connect to, and where we
> can push notifications. We already leveraged that in Heat [1], so that
> you can manipulate a stack and you'll be notified when its status
> changes.
>
> Still, not everyone uses Heat, and under the hood it still polls for
> you. We should be able to use the various notifications that projects
> publish to push similar events. Ceilometer already consumes those
> notifications and try to make them somewhat consumable [2].
>
> The missing piece is something that bridges Ceilometer and Zaqar. I
> wrote a small service [3] which provide a simple API, so that you can
> say "Subscribe to those events and send them to this queue". The data
> flow looks like this:
>
> Service (Nova, Neutron) -> Notification -> Ceilometer -> Bridge ->
> Zaqar -> User.
>
> This way, you'll get a Zaqar message per event, with some filtering
> done by the bridge service. It's far from being ideal, as the
> notification format is a endless topic of conversation, but I hope
> that if we start using it things will move further. I also hope I can
> get some feedback on the approach.

Could you document what makes this different from 
http://docs.openstack.org/developer/aodh/event-alarm.html? (I can think 
of some ways, but it's not clear to me whether a separate service is the 
right thing or if the existing Aodh event alarms can be modified to do 
what we want.)

> One question is whether we want to skip the subscription phase, and
> simply makes all those events available to users at a known place (for
> example, all events in the Zaqar queue events-queue, or maybe all Nova
> events in the compute-events-queue). That pushes filtering to the
> user, but it also simplifies the bootstrap phase.
>
> Anyway, I wanted to send this before the summit so that people
> interested can come talk to me.
>
> Thanks,
>