Fix Galera bootstrap issues

[steventwheeler]

This fixes a few issues I had when bootstrapping a new Galera cluster.

1. A race condition exists when managing the users and database on all nodes. This change restricts these actions to the primary node.
2. I was unable to perform IST without first configuring SST authentication.
3. MariaDB is not restarted on the second and third nodes after configuring Galera.

[HanXHX]

I don't understand your PR.

The good way to use this role: launch playbook on first "bootstrap" node and on a second time launch this role on other nodes.

Did you try to launch this role on all nodes at the same time?

[steventwheeler]

That explains why I struggled getting it up and running. I was running the playbook against all three nodes in my inventory. Is there a reason that two separate runs are preferred?

[HanXHX]

That explains why I struggled getting it up and running. I was running the playbook against all three nodes in my inventory. Is there a reason that two separate runs are preferred?

This Ansible role is very complicated. If you need 'all at the same time', you must put a shared lock system on the first node and wait the bootstrap before executing other tasks...

I don't know how to perform this lock and keep running on the CI 🤣

But I can write a better doc on this setup!

[steventwheeler]

I'm not sure I follow. What do you mean by "shared lock system"? With the changes in this merge request it is able to bootstrap all three nodes in a single run. Also, regarding the CI failure that appears to be a yamllint issue on a line I didn't change.