Project

General

Profile

Bug #2188

mon: mds rm should be harder to break things with

Added by Greg Farnum about 12 years ago. Updated almost 12 years ago.

Status:
Resolved
Priority:
High
Assignee:
-
Category:
-
Target version:
% Done:

0%

Source:
Community (user)
Tags:
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

If you run ceph mds rm 0 on a healthy cluster, it breaks the Monitor's world. I'm uncomfortable with the command existing, but don't want to remove it on my own.
Still, if it stays, it should have:
1) A yes-i-really-mean-it guard
2) Some method of dealing with actually removing a live MDS. Returning an error if it's up? Pulling it out of the up set and mds_info (I think that might force it to shut down, though it leaves the logical around)?
3) Some documentation.

Associated revisions

Revision b50a4c92 (diff)
Added by Sage Weil almost 12 years ago

mon: add safety checks for 'mds rm <gid>' command

- make sure the gid exists
- only remove it if it's inactive (state < 0)

Fixes: #2188
Signed-off-by: Sage Weil <>

History

#1 Updated by Sage Weil almost 12 years ago

  • Status changed from New to Resolved
  • Target version set to v0.47

added checks that the gid exist and that it is not active. with those in place, this is no longer dangerous... all you can do is remove a standby or other stray entry, so i didn't add an extra guard.

docs can come later; this isn't the first thing about the mds to write down.

Also available in: Atom PDF