Question about IPA installation and repos more generally

[verdurin]

My invocation of ipa.yml is failing because it can't find a repo from which to install ipa-client.

Most of the usual repos are set to Ark addresses, and disabled.

I've been looking at the Operations guide that talks about adding repos and packages in image builds - it's not immediately clear whether that applies to running instances too.

If I haven't missed somewhere this is explained, it would be good to hear how this is expected to work. That would be preferable to just overwriting the existing repo config for want of a cleaner approach.
For example, if the cookiecutter config requires an Ark key, it would be good to state that.

(I did add a local Pulp server on the control node, but I see your support for that is still in the experimental section...)

[sjpb]

Apologies for the slow response.

So, this has changed very rapidly over the last couple of months or so, backstory here: https://stackhpc.com/slurm-package-changes.html.

In general, the instructions assume you are using our images with no additional features enabled (because to-date, any other use has had our involvement). Turning on the freeipa support will involve installing packages so yes, will require ark credentials at some point (either directly, to use for image build, or indirectly to allow a local pulp deployment to sync). The other option for testing is simply to overwrite the .repo files yourself to refer to the upstream mirrors and reenable them to use upstream packages. There no built-in support for doing that though, because it ends up in an untested configuration in terms of package configurations.

The docs for the Pulp server are still in the experimental but they do work - the steps required may change though as we get experience doing this. But again, you still need ark credentials to sync it.

Agree the docs should make it clear that enabling any non-default feature which requires package installs currently needs ark credentials.