To upgrade an existing installation of Tungsten Cluster, the new distribution must be downloaded and unpacked, and the included tpm command used to update the installation. The upgrade process implies a small period of downtime for the cluster as the updated versions of the tools are restarted, However the process that takes place should not present as an outage to your applications providing steps when upgrading the connectors are followed carefully. Any downtime is deliberately kept to a minimum, and the cluster should be in the same operational state once the upgrade has finished as it was when the upgrade was started.
During the update process, the cluster will be in MAINTENANCE mode.
This is intentional to prevent unwanted failovers during the process, however it is
important to understand that should the primary fail for genuine reasons NOT
associated with the upgrade, then failover will also not happen at that time.
It is important to ensure clusters are returned to the AUTOMATIC state
as soon as all Maintenance operations are complete and the cluster is stable.
It is NOT advised to perform rolling upgrades of the tungsten software to avoid miscommunication
between components running older/newer versions of the software that may prevent switches/failovers from occuring,
therefore it is recommended to upgrade all nodes in place. The process of the upgrade places the
cluster into MAINTENANCE mode which in itself avoids outages whilst components are restarted,
and allows for a successful upgrade.
From version 7.1.0 onwards, the JGroup libraries were upgraded, this means that when upgrading to any release
to 7.1.0 onwards FROM any release OLDER than 7.1.0 (for example 6.1.0 -> 7.1.4), all nodes must be upgraded
before full cluster communication will be restored. For that reason upgrades to 7.1.0 onwards from an OLDER
release MUST be performed together, ensuring the cluster is only running with a mix of manager versions for as
little time as possible. When upgrading nodes, do NOT SHUN the node otherwise you will not be able to
recover the node into the cluster until all nodes are upgraded, which could result in an outage to your applications.
Additionally, do NOT perform a switch until all nodes are upgraded. This means you should upgrade the master node in
situ. Providing the cluster is in MAINTENANCE, this will not cause an outage and the cluster can still be upgraded
with no visible outage to your applications.