Project

General

Profile

Actions
Copy link

Documentation #61458

closed

CephFS removal not completely documented (or not working as documented)

Added by Voja Molani 11 months ago. Updated 10 months ago.

Status:
Closed
Priority:
Normal
Assignee:
-
Category:
cephadm
Target version:
-
% Done:

0%

Tags:
Backport:
Reviewed:
Affected Versions:
Ceph - v17.2.5
Pull request ID:

Description

After deleting testing CephFS volumes with ceph fs volume rm and then upgrading the cluster from 17.2.5 to 17.2.6 I noticed MDS running for all the already-removed CephFS volumes. Pretty sure they were stopped after the ceph fs volume rm but cannot say for sure about that - they definitely are running now.

As listed in https://tracker.ceph.com/issues/48597 the mds_join_fs exists for all of the already-removed CephFS volumes.
Additionally the data & metadata pools exist for all the already-removed CephFS volumes.

Currently the only documentation that I could find about removing CephFS volumes is at https://docs.ceph.com/en/latest/cephfs/fs-volumes/#fs-volumes which says to simply ceph fs volume rm the volume which should "try to remove" the running MDS as well.

But this documentation seems far from complete as it does not mention that:
  1. The data & metadata pools are not removed.
  2. The mds_join_fs exist after the pools are removed.
  3. The MDS are running (again?) after the pools are removed. (related to #2?)

The documentation makes it seem like you just need ceph fs volume rm and you're good. But that doesn't seem to be the case at all!
The documentation should be improved to at minimum list what are the leftovers after ceph fs volume rm and possibly what to do to clean the leftovers.

Actions
Copy link
 #1

Updated by Voja Molani 11 months ago

Hate that I cannot edit the OP. Anyhow:

The documentation should explain what does the "tries to remove MDS daemons" means. Does "try" in here mean that it might not be able to remove MDS daemons? If it fails then what should one do? Or does "try" here simply mean that if an orchestrator module exists then MDS daemons are removed?
The choice of words is really not good here, it makes it sound like it is not known/certain if the MDS should be removed after a ceph fs volume rm.

Actions
Copy link
 #2

Updated by Voja Molani 11 months ago

And after re-reading the documentation I noticed the sentence "This removes a file system and its data and metadata pools". But none of my .data / .meta pools are actually deleted... so which one is wrong here, the documentation or cephadm?

Actions
Copy link
 #3

Updated by Voja Molani 11 months ago

Creating and removing some more testing CephFS the .data & .meta pools are removed and the volume is removed from mds_join_fs so after a bit of pondering I understood what had happened. The previous volumes were deleted with ceph fs rm and not ceph volume fs rm. I'll just close this then.

Actions
Copy link
 #4

Updated by Redouane Kachach Elhichou 10 months ago

  • Status changed from New to Closed

Please, feel free to reopen if you still consider that there's an issue with docs or code.

Actions
Copy link

Also available in: Atom PDF