Project

General

Profile

Fix #19549

ceph-mgr should respawn when it goes inactive

Added by John Spray 4 months ago. Updated about 2 months ago.

Status:
Resolved
Priority:
High
Assignee:
-
Category:
ceph-mgr
Target version:
-
Start date:
04/07/2017
Due date:
% Done:

0%

Source:
Development
Tags:
Backport:
Reviewed:
Affected Versions:
ceph-qa-suite:
Release:
Needs Doc:
No

Description

When I wrote the Mgr/MgrStandby code originally, I tried to make it tear down the active Mgr instance cleanly when it received a mgrmap that indicated that it was no longer active. That's in contrast to how the MDS does it, by (somewhat brutally) respawning its process.

I've now seen at least one case where it simply isn't possible to reliably tear down the python interpreter (daemon threads http://bugs.python.org/issue231189), so I think in the interests of robustness we should try and do a clean teardown of everything, and then finally, after destroying the Mgr instance, respawn the process.

History

#1 Updated by Sage Weil about 2 months ago

  • Priority changed from Normal to High

#2 Updated by Sage Weil about 2 months ago

  • Status changed from New to Resolved

Also available in: Atom PDF