Chapter 36. Rolling Upgrades

In Red Hat JBoss Data Grid, rolling upgrades permit a cluster to be upgraded from one version to a new version without experiencing any downtime. This allows nodes to be upgraded without the need to restart the application or risk losing data.
In JBoss Data Grid, rolling upgrades can only be performed in Remote Client-Server mode.

Important

When performing a rolling upgrade it is recommended to not update any cache entries in the source cluster, as this may lead to data inconsistency.
Report a bug

36.1. Rolling Upgrades Using Hot Rod

The following process is used to perform rolling upgrades on Red Hat JBoss Data Grid running in Remote Client-Server mode, using Hot Rod. This procedure is designed to upgrade the data grid itself, and does not upgrade the client application.

Important

Ensure that the correct version of the Hot Rod protocol is used with your JBoss Data Grid version. This version must be specified on the client programmatically, and instructions on defining this are found inside the JBoss Data Grid Developer Guide. A list of Hot Rod protocol versions from each release are found below:
  • For JBoss Data Grid 7.0, use Hot Rod protocol version 2.5
  • For JBoss Data Grid 6.6, use Hot Rod protocol version 2.3
  • For JBoss Data Grid 6.5, use Hot Rod protocol version 2.0
  • For JBoss Data Grid 6.4, use Hot Rod protocol version 2.0
  • For JBoss Data Grid 6.3, use Hot Rod protocol version 2.0
  • For JBoss Data Grid 6.2, use Hot Rod protocol version 1.3
  • For JBoss Data Grid 6.1, use Hot Rod protocol version 1.2
Prerequisite

This procedure assumes that a cluster is already configured and running, and that it is using an older version of JBoss Data Grid. This cluster is referred to below as the Source Cluster and the Target Cluster refers to the new cluster to which data will be migrated.

  1. Configure the Target Cluster

    Use either different network settings or a different JGroups cluster name to set the Target Cluster (consisting of nodes with new JBoss Data Grid) apart from the Source Cluster. For each cache, configure a RemoteCacheStore with the following settings:
    1. Ensure that remote-server points to the Source Cluster.
    2. Ensure that the cache name matches the name of the cache on the Source Cluster.
    3. Ensure that hotrod-wrapping is enabled (set to true).
    4. Ensure that purge is disabled (set to false).
    5. Ensure that passivation is disabled (set to false).
    Configure the Target Cluster with a RemoteCacheStore

    Figure 36.1. Configure the Target Cluster with a RemoteCacheStore

    Note

    See the $JDG_HOME/docs/examples/configs/standalone-hotrod-rolling-upgrade.xml file for a full example of the Target Cluster configuration for performing Rolling Upgrades.

  2. Start the Target Cluster

    Start the Target Cluster's nodes. Configure each client to point to the Target Cluster instead of the Source Cluster. Eventually, the Target Cluster handles all requests instead of the Source Cluster. The Target Cluster then lazily loads data from the Source Cluster on demand using the RemoteCacheStore.
    Clients point to the Target Cluster with the Source Cluster as RemoteCacheStore for the Target Cluster.

    Figure 36.2. Clients point to the Target Cluster with the Source Cluster as RemoteCacheStore for the Target Cluster.

  3. Dump the Source Cluster keyset

    When all connections are using the Target Cluster, the keyset on the Source Cluster must be dumped. This can be done using either JMX or the CLI:

    • JMX

      Invoke the recordKnownGlobalKeyset operation on the RollingUpgradeManager MBean on the Source Cluster for every cache that must be migrated.

    • CLI

      Invoke the upgrade --dumpkeys command on the Source Cluster for every cache that must be migrated, or use the --all switch to dump all caches in the cluster.

  4. Fetch remaining data from the Source Cluster

    The Target Cluster fetches all remaining data from the Source Cluster. Again, this can be done using either JMX or CLI:

    • JMX

      Invoke the synchronizeData operation and specify the hotrod parameter on the RollingUpgradeManager MBean on the Target Cluster for every cache that must be migrated.

    • CLI

      Invoke the upgrade --synchronize=hotrod command on the Target Cluster for every cache that must be migrated, or use the --all switch to synchronize all caches in the cluster.

  5. Disabling the RemoteCacheStore

    Once the Target Cluster has obtained all data from the Source Cluster, the RemoteCacheStore on the Target Cluster must be disabled. This can be done as follows:

    • JMX

      Invoke the disconnectSource operation specifying the hotrod parameter on the RollingUpgradeManager MBean on the Target Cluster.

    • CLI

      Invoke the upgrade --disconnectsource=hotrod command on the Target Cluster.

  6. Decommission the Source Cluster

    As a final step, decommission the Source Cluster.
Report a bug