There are both proprietary and open source high-availability software stacks. Examples of proprietary ones are Solaris Cluster (sometimes called Sun Cluster or SunCluster), SteelEye LifeKeeper, Evidian SafeKit, and others. We don't elaborate on them in detail in this book.
Cluster software usually contains two distinct layers: the transport layer and the cluster management layer. The management layer is responsible for starting and stopping services on cluster nodes. The service and health information is communicated via the transport layer.
Initially, there were two widely known open source high-availability software stacks, called OpenAIS and Linux-HA. These were mutually incompatible and both had their strengths and weaknesses. Later, the two developer communities joined forces, starting with an internal announcement on the Linux-HA users mailing list on December 7, 2007. The management layer of Linux-HA (called CRM at the time) was to be split out to support both the original Linux-HA and OpenAIS' transport layers. There was another public announcement in 2008 at the Ottawa Linux Symposium, where the OpenAIS transport layer was to split up to support the new, common management layer better. The previous, somewhat monolithic structure of both stacks became lighter, compatible, and interchangeable.
The first stable versions implementing the joint effort are:
- Heartbeat Version 3.0.2: February 1, 2010
- Corosync Version 1.0.0: July 8, 2009
- Pacemaker Version 0.6: January 16, 2008
OpenAIS was the first to implement the Service Availability Forum (www.saforum.org) specification. It was a comprehensive cluster software stack but also a complex one. At the time, Corosync was the synonym of OpenAIS. In 2008, the project developers announced the joint development at the Ottawa Linux Symposium and the result was that the software was refactored, its transport layer became Corosync and the OpenAIS part now only contains the SAForum API. However, at the time of writing this, the SAForum site only lists OpenSAF (opensaf.org) and OpenHPI (www.openhpi.org) as the implementers of their specification.
The Linux-HA stack (www.linux-ha.org) started out as a simple cluster implementation, to provide an easy way to set up a two-computer cluster. The transport layer was called Heartbeat, but it also became a synonym of Linux-HA since the upper layer, the cluster manager didn't have a specific name. The simplicity of Heartbeat Version 1.x was considered a weakness after some time and the cluster management layer was rewritten and became what was called the Cluster Resource Manager (CRM). After the joint OpenAIS and Linux-HA development, the transport layer became a separate piece of software, keeping the Heartbeat name and the CRM was factored out and was renamed to Pacemaker. Currently, Pacemaker supports both projects' transport layers (Heartbeat and Corosync) and provides a common cluster manager layer. For a few applications, it even needs components from the OpenAIS manager layer. The current homepage for Heartbeat and Pacemaker is www.clusterlabs.org.