OpenStack Summit May 2015 Vancouver

Service high-availability is a core concern for any operator running, or intending to run, a production OpenStack cluster. Current "best practices" call for the deployment of a full cluster stack (generally Corosync + Pacemaker) and / or load-balancers (generally HAProxy) to manage service state and failover. While this is appropriate for a subset of OpenStack services it unnecessarily adds complexity to many others.

 

In this discussion I'll help fellow operators exploit the hard work and knowledge of our

network engineering brethren by providing an overview of the 'anycast' routing methodology

and supporting components (e.g. BGP, OSPF, Quagga, etc.), and discussing it's implication on providing service high-availability in an OpenStack environment.