Dominic Williams

Occasionally useful posts about RIAs, Web scale computing & miscellanea

Posts Tagged ‘rolling upgrade

10 steps to upgrade a Cassandra node

with 2 comments

The following steps might prove interesting to those with a new Cassandra cluster who are wondering how to upgrade, or to those investigating Cassandra who are interested in what an online upgrade looks like. The good news for those in production, is that if your cluster has a replication factor (RF) of at least 3, read and write operations (with the exception of rarely used ConsistencyLevel.All operations) will be able to continue uninterrupted throughout the upgrade process i.e. you don’t need to put up a “down for maintenance” sign!!

This process should be performed on every node in the cluster in sequence to effect a rolling upgrade. Once you are comfortable with the process, and the number of nodes in your cluster starts to climb, you will want to script much of it.

1. Open a ssh window to the Cassandra node.
$ ssh

2. Visit The homepage has a link to the latest tar.gz package download. Copy the url of that link

3. Change folder to /opt and then download the upgrade package.
$ cd /opt
$ wget…cassandra/0.7.4/apache-cassandra-0.7.4-bin.tar.gz

4. Unpack the package and enter its configuration folder
$ tar xvf apache-cassandra-0.7.4-bin.tar.gz
$ cd /opt/apache-cassandra-0.7.4/conf

5. Backup the default configuration (sometimes the format changes). Copy in the existing configuration.
$ mv cassandra.yaml cassandra.yaml.bak
$ cp /opt/cassandra/conf/cassandra.yaml .

6. Edit the environment configuration script. This often changes so don’t copy in the existing script. You need to configure heap size to be ~50% RAM on server and new heap size to be 100MB * no. CPUs
$ vim cassandra.yaml
…then, for example for server with 6GB RAM and 6CPUs

7. Create a link to jna.jar from Cassandra’s lib folder (JNA or “Java Native Access” enables Cassandra to directly access native libraries. Without this capability, certain operations such as snapshotting using nodetool are more likely to fail with OOM).
$ cd /opt/apache-cassandra-0.7.4/lib
$ ln -s /usr/share/java/jna.jar .

8. Shutdown the Cassandra software running on the current node. This will not interrupt the cluster in any way.
$ pkill -f ‘java.*cassandra’

9. Change to the “opt” folder and change the soft link the Cassandra distribution
$ cd /opt
$ rm cassandra
$ ln -s /opt/apache-cassandra-0.7.4 cassandra

10. Restart Cassandra
$ cd /opt/cassandra/bin
$ ./cassandra

After restarting Cassandra, you should watch the startup messages to ensure the node comes up without difficulty.

Written by dominicwilliams

March 26, 2011 at 9:23 pm