Project

General

Profile

Actions
Copy link

Bug #18750

open

handle_pg_remove: pg_map_lock held for write when taking pg_lock

Added by Josh Durgin about 7 years ago. Updated almost 7 years ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
Performance/Resource Usage
Target version:
-
% Done:

0%

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

Description

This could block the fast dispatch path, since fast dispatch takes pg_map_lock for read, so this makes it possibly block on the pg_lock.

Check for other instances of pg_map_lock being taken for write as well.

Actions
Copy link
 #1

Updated by Greg Farnum almost 7 years ago

  • Project changed from Ceph to RADOS
  • Category changed from OSD to Performance/Resource Usage
  • Priority changed from High to Normal
  • Component(RADOS) OSD added

I think I remember this one and it wasn't really feasible to fix (at the time). If doing code inspection you'll want to focus on the big PG state change operations (like this one is removing it entirely), which I'm not sure is worth our time right now...

Actions
Copy link

Also available in: Atom PDF