cephadm should use the same image reference across the cluster
The documentation has this warning (from https://docs.ceph.com/docs/octopus/install/containers/#containers):
Important Using the :latest tag is discouraged. If you use the :latest tag, there is no guarantee that the same image will be on each of your hosts. Under these conditions, upgrades might not work properly. Remember that :latest is a relative tag, and a moving target.
Instead of the :latest tag, use explicit tags or image IDs. For example:
podman pull ceph/ceph:v15.2.0
This is an administrative burden on the user to always look up and bump the number, and it's an additional human action that can go wrong.
ceph-ansible solved this by resolving the image's floating tag to a sha256 ref one time, and then deploying that same ref to all nodes. cephadm should be able to do something similar.