Overview

AxonOps provides a rolling restart feature for Cassandra.

The feature is accessible via Operations > Rolling Restart.

The AxonOps user requires permissions to stop and start the Cassandra service. Add the AxonOps user to sudoers, for example:

/etc/sudoers.d/axonops

axonops ALL=NOPASSWD: /sbin/service cassandra *, /usr/bin/systemctl * cassandra*

You can start an immediate rolling restart or schedule it.

Node health check

Before stopping each node, AxonOps checks that all nodes in the cluster are up. If any node is found to be down at that point, the rolling restart fails immediately and does not proceed to stop the next node. This prevents the cluster from losing an additional node when it is already in a degraded state.

The script field lets you customize the predefined script executed by axon-agent during the restart process.

You can also specify different degrees of parallelism for the restart: DC, Rack, and Node.

For example, to restart one entire rack at once across the cluster, set a large Node parallelism (greater than the number of nodes in the rack, for example 999).

DC parallelism: 1
Rack parallelism: 1
Node parallelism: 999

To restart one entire rack across each DC:

DC parallelism: 999
Rack parallelism: 1
Node parallelism: 999