Impact of Pacemaker Failover Configuration on Mean Time to Recovery for Small Cloud Clusters

In cloud environments High Availability characteristics are established by the usage of failover software (like e.g. HAProxy, Keepalive or Pacemaker). Though these tools enable automatic recovery of cloud services from outages, the recovery can still be very slow if it is not configured adequately. In this paper we developed a "Recovery Time Test" to determine if recovery time depends on configuration of the failover software and how recovery time depends on configuration settings. Another goal of the Recovery Time Test is to determine the factor by which recovery time can be decreased by a given configuration. As proof of concept, we applied the Recovery Time Test to an OpenStack cloud environment which is controlled by the Pacemaker failover software. Pacemaker mean recovery time can take a value between 110 and 160 seconds, if the tool is configured badly. We found that with a proper configuration Pacemaker mean recovery time can be reduced significantly to a value between 15 and 20 seconds.