Would it be feasible to have a playbook which migrates a baremetal cluster to containers and, in the process, upgrades Ceph from Jewel to Luminous?

This sounds terrifying to implement. I can't fathom how, if implemented, we would attempt to test such a thing. The burden of support and maintenance would be tremendous.

Most of the "infrastructure playbooks" are a source of concern, and adding something like what is being proposed here would not improve that situation.

This sounds more like an administrator task rather than a ceph-ansible responsibility

(In reply to Alfredo Deza from comment #4)
> This sounds terrifying to implement. I can't fathom how, if implemented, we
> would attempt to test such a thing. The burden of support and maintenance
> would be tremendous.
> 
> Most of the "infrastructure playbooks" are a source of concern, and adding
> something like what is being proposed here would not improve that situation.
> 
> This sounds more like an administrator task rather than a ceph-ansible
> responsibility

Undestood. Note that in OSP we consume already two infrastructure playbooks: switch-from-non-containerized and rolling_update.

We could just run the two, sequentially, to get cluster into the wanted state but that implies downloading a 2.x container image to do the initial migration to containers and then a 3.x container image to do the upgrade.

Do you think that'd still be a better (safer maybe?) approach?

I cant' tell how will that play out in a production environment. I'm still concerned on the wild permutations of environments and attempting to mangle them all with a single playbook.

In short: I don't know how safe (if at all) is that approach.

For now, this is pretty well handled by switch-from-non-containerized, we also have CI coverage for this. The mix of switch-from-non-containerized and rolling_update will be required.

*** This bug has been marked as a duplicate of bug 1488566 ***