Idempotent incus module

[breml]

While working with my home lab with actual hardware I found the existing conditions, which decide, if a particular task is executed or not, problematic, in particular for the incus part. There the condition for a good number of tasks is install_deb.changed or install_rpm.changed.

In my tests, the installation of the .deb package was already successful in a previous run, but some of the later incus command based tasks failed. In the consecutive runs, these steps have been skipped, since the .deb package was already installed and therefore the condition was not met.

For the ansible playbook to become more reliable and "self healing", these tasks need to be rewritten in an idempotent way.

[stgraber]

It's pretty unlikely that we'd ever get full support for this.

Basic configuration keys, we can read and update, but say you change the type of a storage pool in the config, there's no good way to handle that. Same goes for networks.

As soon as the cluster is deployed, you're likely to be moving over to other management tools like Terraform to create objects on the cluster, which then causes dependency chains that the Ansible playbooks won't be able to cleanly undo/redo to apply a config change.

What we could do is key off a user.incus-deploy or something config key in the server configuration and run the playbook steps if that key isn't set, possibly using various values for that key to know how far we got with initialization.

It wouldn't be idempotent but it would allow resuming when stopped/aborted.

[ykazakov]

I wish too the conditions were a little more robust.
This is a very long play, with many possibilities of failures, e.g., due to network problems or other problems not caused by mis-configuration. For example, in one of my runs the deb package failed to install on one of the nodes since there was unattended-updates process running. Upon the second run, the tasks that are triggered by install_deb.changed did not run on the nodes on which the package was installed in the previous run. So the only solution now is wipe everyithing and start from scratch.