Project

General

Profile

Actions

Bug #22165

closed

split pg not actually created, gets stuck in state unknown

Added by Sage Weil over 6 years ago. Updated over 5 years ago.

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

0%

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

Description

pg is in state unknown.

the primary shows lots of

remote/smithi133/log/ceph-osd.7.log.gz:2017-11-19 17:18:36.048591 7ff81545b700 20 osd.7 op_wq(6) _process 6.16 item PGQueueable(0x55b1f2008280 prio 63 cost 4194304 e302) queued, waiting_for_pg

but the pg is never created. it comes into being via split but is not instantiated on the new primary for some reason..

/a/sage-2017-11-19_15:21:03-rados-wip-sage2-testing-2017-11-18-2035-distro-basic-smithi/1865695

Actions #1

Updated by Sage Weil over 6 years ago

- pg 6.6 was never created yet
- osd went down
- split
- osd comes up
- osd gets pg_create on 6.6
- does not process a split, does not get pg_create for 6.16

Actions #2

Updated by Greg Farnum over 6 years ago

  • Subject changed from split pg stuck in state unknown (not created) to split pg not actually created, gets stuck in state unknown

Two solutions. Sage thinks it wouldn't be hard to make create code account for splits that need to be processed.

Slightly simpler option is to make monitor refuse to process splits when the pool has PGs which aren't created yet.

Actions #3

Updated by Sage Weil about 6 years ago

  • Status changed from In Progress to Fix Under Review
Actions #4

Updated by Sage Weil over 5 years ago

  • Status changed from Fix Under Review to Resolved

by commit fdfc5c64

Actions

Also available in: Atom PDF