Performing a UCS and Cassandra Upgrade
To perform an upgrade:
- Stop UCS clients.
- Stop all UCS instances.
- Stop all Cassandra instances.
- Perform upgrade on all UCS and Cassandra nodes.
- Start all Cassandra nodes first, then UCS instances.
- Start UCS clients.
Performing a Rolling Upgrade
The rolling upgrade procedure allows you to upgrade UCS instances without service interruption.
- First, each Cassandra instance is sequentially stopped, upgraded, and restarted (Cassandra Rolling Upgrade).
- Second, each UCS (starting from the UCS master, which has schema/manage=true) is sequentially stopped, upgraded, and restarted (UCS Rolling Upgrade).
In order to perform a successful rolling upgrade a Cassandra Quorum must be maintained during the whole procedure.
In the Cassandra cluster with N nodes a Quorum state is achieved when N/2 + 1 replicas are active. The number of replicas is set by the replication-factor set in UCS options.
This means that:
For the first step (Cassandra Rolling Upgrade):
- Cassandra cluster has more than one node (4 is recommended).
- N/2 + 1 nodes are active. For a 4-node cluster, 3 nodes must be active.
- Only one node must be stopped at a time.
- Once the node is upgraded and started, it will request the missed changes from the other nodes.
This process is repeated for as many Cassandra nodes as you have in your Cassandra cluster.
For the second step (UCS Rolling Upgrade):
- From the UCS cluster, the UCS master should be the first to be stopped, upgraded, and restarted.
- Then once the UCS master is initialized, the next UCS can be upgraded.
This process is repeated for as many UCS nodes as you have in your UCS cluster.