Loading…
This event has ended. View the official site or create your own event → Check it out
This event has ended. Create your own
View analytic
Wednesday, May 14 • 9:00am - 9:40am
Ceilometer agents repartition

Sign up or log in to save this to your schedule and see who's attending!

The ceilometer central agent is responsible for polling a lot of different meters from different OpenStack components. To avoid duplication, this is done by a single central agent, which induces a single point of failure.

At the Icehouse summit, we discussed a plan about to address that. Progress has since been made on an underlying co-ordination protocol facade in the form of tooz[1], but little headway has occurred on addressing this SPoF within ceilometer.

The priority of addressing this architectural defect has been further elevated by the central agent assuming responsibility for hardware polling of nodes over SNMP.

The session will be about how we're going to tackle this definitely in Juno, allowing for the central agent to be scaled out horizontally, with the workload partitioned across the pool of running agents.

Also an unrelated optimization that could possibly be accommodated in the resource discovery layer will be discussed. Specifically we could avoid imposing undue load on nova-api from the compute agent polling, by first calling virt.inspector.inspect_instances() and only hitting nova-api if the hypervisor reports a previously unknown instance (modulo resource-metadata staleness)

[1] https://github.com/stackforge/tooz

(Session proposed by Julien Danjou)


Wednesday May 14, 2014 9:00am - 9:40am
B305

Attendees (37)