Documentation feedback: --cluster-init

[knweiss]

The Upgrades section of the documentation currently reads "To upgrade K3s from an older version you can re-run the installation script using the same flags".

However, when we set up a k3s "HA with embedded etcd db" configuration we still wondered if it is really safe to specify --cluster-init for the upgrade runs, too. The worry was that this would reset ("initizalize") the etcd database.

[ChristianCiach]

It is not only safe, it is necessary. I've done it countless times.

In fact, if you run k3s using a systemd unit (which is the case after installing k3s using the installation script), k3s gets started with --cluster-init every time anyway.

You can also use --cluster-init if your control plane (or even your whole cluster) only contains a single node.

For these reasons I think that --cluster-init is a confusing name for this option. I think it should be named something like --embedded-etcd.

[brandond]

--cluster-init initializes a new etcd cluster if no etcd data is found on disk. The presence of an etcd datastore on disk supersedes all other options, ensuring that the cluster does not get reinitialized if already configured.

[gunzy83]

Just for my understanding: Does this mean that on upgrade the installer could just be run without the --cluster-init option and it would automatically pick up the etcd files on disk and rejoin the cluster?

[brandond]

Yep

[gunzy83]

Thanks!