Chapter 3. Performing Rolling Upgrades for Data Grid Servers

Perform rolling upgrades of your Data Grid clusters to change between versions without downtime or data loss. Rolling upgrades migrate both your Data Grid servers and your data to the target version over Hot Rod.

3.1. Setting Up Target Clusters

Create a cluster that runs the target Data Grid version and uses a remote cache store to load data from the source cluster.

Prerequisites

  • Install a Data Grid cluster with the target upgrade version.
Important

Ensure the network properties for the target cluster do not overlap with those for the source cluster. You should specify unique names for the target and source clusters in the JGroups transport configuration. Depending on your environment you can also use different network interfaces and specify port offsets to keep the target and source clusters separate.

Procedure

  1. Add a RemoteCacheStore on the target cluster for each cache you want to migrate from the source cluster.

    Remote cache stores use the Hot Rod protocol to retrieve data from remote Data Grid clusters. When you add the remote cache store to the target cluster, it can lazily load data from the source cluster to handle client requests.

  2. Switch clients over to the target cluster so it starts handling all requests.

    1. Update client configuration with the location of the target cluster.
    2. Restart clients.

3.1.1. Remote Cache Stores for Rolling Upgrades

You must use specific remote cache store configuration to perform rolling upgrades, as follows: 

<persistence passivation="false"> 1
   <remote-store xmlns="urn:infinispan:config:store:remote:10.1"
                 cache="myDistCache" 2
                 protocol-version="2.5" 3
                 hotrod-wrapping="true" 4
                 raw-values="true"> 5
      <remote-server host="127.0.0.1" port="11222"/> 6
   </remote-store>
</persistence>
1
Disables passivation. Remote cache stores for rolling upgrades must disable passivation.
2
Matches the name of a cache in the source cluster. Target clusters load data from this cache using the remote cache store.
3
Matches the Hot Rod protocol version of the source cluster. 2.5 is the minimum version and is suitable for any upgrade paths. You do not need to set another Hot Rod version.
4
Ensures that entries are wrapped in a suitable format for the Hot Rod protocol.
5
Stores data in the remote cache store in raw format. This ensures that clients can use data directly from the remote cache store.
6
Points to the location of the source cluster.

Reference

3.2. Synchronizing Data to Target Clusters

When your target cluster is running and handling client requests using a remote cache store to load data on demand, you can synchronize data from the source cluster to the target cluster.

This operation reads data from the source cluster and writes it to the target cluster. Data migrates to all nodes in the target cluster in parallel, with each node receiving a subset of the data. You must perform the synchronization for each cache in your Data Grid configuration.

Procedure

  1. Start the synchronization operation for each cache in your Data Grid configuration that you want to migrate to the target cluster.

    Use the Data Grid REST API and invoke GET requests with the ?action=sync- data parameter. For example, to synchronize data in a cache named "myCache" from a source cluster to a target cluster, do the following: 

    GET /v2/caches/myCache?action=sync-data

    When the operation completes, Data Grid responds with the total number of entries copied to the target cluster.

    Alternatively, you can use JMX by invoking synchronizeData(migratorName=hotrod) on the RollingUpgradeManager MBean.

  2. Disconnect each node in the target cluster from the source cluster.

    For example, to disconnect the "myCache" cache from the source cluster, invoke the following GET request: 

    GET /v2/caches/myCache?action=disconnect-source

    To use JMX, invoke disconnectSource(migratorName=hotrod) on the RollingUpgradeManager MBean.

Next steps

After you synchronize all data from the source cluster, the rolling upgrade process is complete. You can now decommission the source cluster.