Currently, the data pool names always have the filesystem name prefix and the customization of the pool name is only after the prefix (see generateDataPoolNames()). For consistency, seems like the metadata pool name should also have the filesystem name prefix. Or is there any need to not use the filesystem name as a prefix?

You are right here. The documentation has resulted in a misunderstanding as it says [...] a pool name can be specified with the name field to override the default generated name; see more below.. But further below it describes that the pool name is generated anyway of what is specified. It only appends it to the filesystem name combined with an -.

Do you see any possibility and does it sound reasonable for your to add a flag to really be able to specify the pool names manually? I could add a flag like generatePoolNames default to true to pools [1].

[1] https://rook.github.io/docs/rook/latest-release/CRDs/Shared-Filesystem/ceph-filesystem-crd/#pools

The idea behind this request is that I come with migration of exisiting resources in mind. Ceph documents that MDS uses pool names like cephfs_metadata and cephfs_data. Currently there is no way in Rook to use these kinds of names as they always get prefixed including - in it's name.

See https://docs.ceph.com/en/latest/cephfs/createfs/#creating-pools

Can you say more about your migration? Rook has never had the goal to take over a ceph cluster. There are too many different assumptions in the deployments of the components such as mons, osds, the naming convention of pools, etc.

If we need to ability to set any pool name for cephfs that will be easy enough if it's needed, I'm more concerned about the attempt to migrate which is a bigger goal.

The request is coming from #14971. A third-party tool provides support on doing an migration with minimal downtime just touching Ceph daemons and replacing them with Rook based ones. A blocker we had is that most traditional Ceph clusters use e.g. cephfs_metadata as documented upstream.

Ceph configuration already stored at e.g. "ceph-mon" daemons are not touched and do work after migration as long as Rook does not try to change these settings. We haven't encountered severe issues in the migration process but our migration scenario required had been pretty straight forward so far.

I've added an boolean now to decide if metadata and data pool names are used as is or default behavior is used.

Interesting to hear the migration is actually working. At some point, more comments on #14971 would be helpful to understand these steps more, or perhaps it's already documented in that rookify project.

Interesting to hear the migration is actually working. At some point, more comments on #14971 would be helpful to understand these steps more, or perhaps it's already documented in that rookify project.