25.11 The Non-Rolling Patch Upgrade

While the current patch strategy of performing a Rolling Upgrade is designed to provide the greatest amount of availability, many systems managers have indicated that this high availability option is not utilized for mission critical services. Although the cluster is up and available, as soon as the installation of a Patch Kit starts, the cluster is considered to be down and therefore unavailable. Understandably, systems managers want a solution to apply patches that is simpler, quicker to implement, and minimizes the number of reboots.

The cluster engineers at HP have come up with an inventive solution to this new requirement. Instead of performing a Rolling Upgrade for a Patch Kit, they will provide new functionality to allow for a Non-Rolling Patch Upgrade.

How would this work? The dupatch program will be modified to provide the option of patching the entire cluster as a single unit. dupatch will be run on one member of the cluster and once it completes the management of the patches, it will issue a command to the other cluster members to perform the patch propagation, the patch configuration, and the building of the individual kernels for each of the cluster members. This also means that while the Non-Rolling Patch Upgrade is in progress, there should be no users or user applications running on the cluster.

One note about the Non-Rolling Patch Upgrade process: because this uses the Event Management (EVM) subsystem^[8] within Tru64 UNIX, it is necessary that all cluster nodes be in multi-user mode with network services up and available.

This all sounds like a nice future feature right? As of this writing, preliminary testing has been quite impressive with the Non-Rolling Patch Upgrade. Using this new feature, the time to patch a three-node cluster on Tru64 UNIX version 5.1 and TruCluster Server version 5.1 is approximately 36 minutes. An eight-node cluster took 1 hour 20 minutes.

When will this feature be available and on which versions of TruCluster Server? This new feature is scheduled to be introduced starting with Patch Kit 2 for Tru64 UNIX version 5.1A/TruCluster Server 5.1A in late Spring 2002. It is not known if HP will provide this same functionality with later Patch Kits for V5.0A and V5.1.

^[8]For more information on the Event Management (EVM) subsystem, please see Chapter 8 – the Event Manager.