A Rolling Upgrade is an upgrade which proceeds through a cluster one node at a time. The fundamental rule of a Rolling Upgrade is that the cluster as a whole is never shut down. That is, even though individual machines are rebooted, or services migrated from one machine to another, the cluster itself never stops providing service.
Small service interruptions may be observed, but since the cluster as a whole never shuts down, service remains basically uninterrupted throughout the upgrade process. During a Rolling Upgrade, all resources remain at all times under the supervision of the LocalResourceManager[1] and the ClusterResourceManager[2].
This is the model which Heartbeat[3] has used for all its upgrades up until the present time.
This is several steps short of the more complex and risky TransparentUpgrade[4] -- in which no service interruption is visible at all, but the resources run for period of time not under supervision.
| [1] | http://www.linux-ha.org/LocalResourceManager |
| [2] | http://www.linux-ha.org/ClusterResourceManager |
| [3] | http://www.linux-ha.org/Heartbeat |
| [4] | http://www.linux-ha.org/TransparentUpgrade |
This information provided courtesy of the Linux-HA project at http://linux-ha.org/