VERITAS Volume Manager 3.5 Administrator's Guide (September 2002)

Upgrading Cluster Functionality
258 VERITAS Volume Manager Administrator’s Guide
Cluster Shutdown
If all nodes leave a cluster, shared volumes must be recovered when the cluster is next
started if the last node did not leave cleanly, or if resynchronization from previous nodes
leaving uncleanly is incomplete.
Upgrading Cluster Functionality
The rolling upgrade feature allows you to upgrade the version of VxVM running in a
cluster without shutting down the entire cluster. To install the new version of VxVM
runningon a cluster, makeonenode leavethecluster, upgrade it, andthen join itbackinto
the cluster. This operation is repeated for each node in the cluster.
Each VERITAS Volume Manager release starting with Release 3.1 has a cluster protocol
version number associated with it. The cluster protocol version is not the same as the
releasenumber or the disk groupversion number. Thecluster protocol version isstored in
the /etc/vx/volboot file. During a new installation of VxVM, the vxdctl init
command creates the volboot file and sets the cluster protocol version to the highest
supported version.
Each new VERITAS Volume Manager release supports two cluster protocol versions. The
lower version number corresponds to a previous VERITAS Volume Manager release. This
has a fixed set of features and communication protocols. The higher version number
corresponds to the new release of VxVM which has a new set of these features. If the new
release of VxVM does not have any functional or protocol changes, but only bug fixes or
minor changes, the cluster protocol version remains unchanged. In this case, the cluster
protocol version does not need to be upgraded.
During a rolling upgrade, each node must be shut down and the VERITAS Volume
Manager release with the latest cluster protocol version must be installed. All nodes that
have the new release of VxVM continue to use the lower level version. A slave node that
has the new cluster protocol version installed tries to join the cluster. If the new cluster
protocol version is not in use on the master node, it rejects the join and provides the
current cluster protocol version to the slave node. The slave retries the join with the
cluster protocol version provided by the master node. If the join fails at this point, the
cluster protocol version on the master node is out of range of the protocol versions
supportedby thejoiningslave. In sucha situation, youmustupgrade theremainderof the
cluster through each intermediate release of VxVM to reach the latest supported cluster
protocol version.
Once you have installed the new release on all nodes, run the vxdctl upgrade
command on the master node to switch the cluster to the higher cluster protocol version.
See “Upgrading the Cluster Protocol Version” on page 268 for more information.