My name is David Vossel. I’m a core developer of the Pacemaker project and author of Pacemaker Remote. I want to share our success with scaling the Pacemaker cluster resource manager to meet the requirements necessary automate recovery of OpenStack controller and compute nodes.
OpenStack’s architecture consists of a set of distributed components capable of scaling horizontally across 100s and even 1000s of nodes. By introducing Pacemaker we are essentially giving the cluster the equivalent of a PID 1, a distributed init system capable of reliably monitoring and recovering all the OpenStack components across the entire cluster.
At this point those of you with previous High Availability experience might be scratching your head in confusion, and rightfully so. Traditionally High Performance Clusters and High Availability clusters have been mutually exclusive. From this standpoint the concept of a "PID 1" within a scalable system seems impractical, but times are changing. High Availability no longer means what it used to mean. Recent architecture improvements to Pacemaker’s ability to scale have brought these two cluster concepts into the same space. As a result, we now have a cluster resource manager capable of managing OpenStack at scale. Now what was previously thought to be impractical is quickly becoming an essential element in what will drive the long term success of OpenStack.
