When upgrading an Artifactory HA installation, there should not be a need for full shutdown of the cluster.
The zero-downtime upgrade process will be as follow:
- Starting from a cluster running with one primary master and many slaves.
- Executing an upgrade of the primary master, will generate a downtime of the master, but the load balancer should redirect all queries to the slaves.
- When the master starts it updates the configuration and the DB schema then send a reload configuration message to the slaves.
- All slaves receiving the message will compare the version to their actual running version, and if it does not match they shutdown gracefully.
- Then running the upgrade procedure on all slaves will bring the whole cluster to the latest version.