Project

General

Profile

Actions
Copy link

Bug #5226

closed

Some PG stay in "incomplete" state

Added by Olivier Bonvalet almost 11 years ago. Updated over 10 years ago.

Status:
Won't Fix
Priority:
Normal
Assignee:
-
Category:
-
Target version:
-
% Done:

0%

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

Description

Hi,

With bobtail I first loose the OSD.25 : the OSD process was crashing, and when its data are ballanced on other OSD (because of reweigth or because of "out" state on OSD.25) they also crash. So, I choose to set the cluster in "noout", waiting for a fix (yes, I should open a bug for that...). I also try to mark it as "lost", and format (mkfs.xfs) without any success. And since cuttlefish I can't start this OSD anymore. It's down, out, and lost.

Then I loose the OSD.19 : the hard disk is dead, unable to read or write any data on it. So, I mark it as "lost" and replace the disk, the OSD.19 is now running.

But : I have 2 pool which use only 2 replica, so I loose the PG which were common to OSD.19 and OSD.25.
In fact I have 15 PG which stay in "incomplete" state, and I don't know how to recover for that. And 14 of thoses 15 PG are related to OSD.19.

Files

Download all files
osd.19.extract.log.gz (19.4 MB) osd.19.extract.log.gz OSD.19 logs Olivier Bonvalet, 05/31/2013 04:21 PM
pg-query.txt (453 KB) pg-query.txt result of "ceph pg 4.0 query" Olivier Bonvalet, 06/03/2013 02:55 PM
Actions
Copy link

Also available in: Atom PDF