Project

General

Profile

Actions
Copy link

Bug #10791

closed

ceph mds deactivate sometimes fails silently

Added by John Spray about 9 years ago. Updated about 9 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
Greg Farnum
Category:
-
Target version:
-
% Done:

0%

Source:
other
Tags:
Backport:
Regression:
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(FS):
Labels (FS):
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

I suspect this is due to the targeted MDS sending a beacon at the same moment as the deactivate, which flips the pending_mdsmap state back from stopping into active.

Happens fairly frequently when doing a "ceph mds set max_mds 1; ceph mds deactivate 1" on a 2-mds vstart cluster.

Files

logs.tar.gz (551 KB) logs.tar.gz John Spray, 02/07/2015 02:01 PM
Actions
Copy link
 #1

Updated by John Spray about 9 years ago

  • Description updated (diff)
Actions
Copy link
 #2

Updated by Greg Farnum about 9 years ago

  • Status changed from New to In Progress
  • Assignee set to Greg Farnum

Yep, that's a pretty clear race in the source.

Actions
Copy link
 #3

Updated by Greg Farnum about 9 years ago

  • Status changed from In Progress to Fix Under Review

I think what we really want to do is have the state_seq go in both directions and force the MDS to acknowledge the transition to STATE_STOPPING before we accept any new beacons from them, but that's not really feasible in the Hammer timeframe. So here's a patch that just doesn't allow any transitions except to stopping: https://github.com/ceph/ceph/pull/3683

Actions
Copy link
 #4

Updated by Greg Farnum about 9 years ago

  • Status changed from Fix Under Review to Resolved
Actions
Copy link

Also available in: Atom PDF