Monitoring performance of OpenStack services

Registered by Sandy Walsh on 2012-08-27

OpenStack uses a queuing service for its RPC transport. Nova performance can suffer due to the "inflight time" where the call (and response) is held in the queuing service. We propose a service that periodically injects negligable overhead messages into the RPC queues and computes the inflight time as well as greenthread overhead. The collected data may be sent to whatever graphing / monitoring program the administrators have selected, but there will be default support for the OpenStack logger and statsd.

Blueprint information

Status:
Complete
Approver:
Russell Bryant
Priority:
Low
Drafter:
Sandy Walsh
Direction:
Approved
Assignee:
Sandy Walsh
Definition:
Superseded
Series goal:
None
Implementation:
Not started
Milestone target:
None
Completed by
Russell Bryant on 2013-04-25

Related branches

Sprints

Whiteboard

This seems like a reasonable set of openstack internal performance data that would be useful. I'd like to see as much of this go into openstack-common as possible so that it could be used by all projects using rpc. It may be worth an openstack-dev thread on the design up front. --russellb

Actually, I think I have a way to get the same results using StackTach ... going to investigate shortly. --sandy-walsh

Superceded by: https://blueprints.launchpad.net/ceilometer/+spec/stacktach-integration
(unable to link via launchpad)

(?)

Work Items

This blueprint contains Public information 
Everyone can see this information.