Ceph

Hi,

After migrating an existing cluster to cephadm, I decided to remove the Ceph packages from the cluster nodes, except cephadm. As a result, after the deployment of this change to the cluster, I lost the /etc/ceph directory on all nodes and when I reinstalled ceph-common to have the ceph command available on the nodes, I had a hard time to rebuild the keyring (in fact it is pretty simple once you know how to do it but finding the info took me some time). Looking a ceph-common RPM uninstall script, I saw that this script does a "rm -Rf /etc/ceph" rather than renaming it (or backuping it before removing it). IMO it is unexpected.

I suggest improving the uninstall script to use the common approach of renaming (something like /etc/ceph.rpmsave for example) or backuping the directory.

I had the problem with Octopus but check that Pacific ceph-common RPM has the same uninstall script.

Michel