Actions
Bug #8675
closedUnnecessary remapping/backfilling?
Status:
Won't Fix
Priority:
Low
Assignee:
-
Category:
-
Target version:
-
% Done:
0%
Source:
Community (user)
Tags:
Backport:
Regression:
No
Severity:
4 - irritation
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(RADOS):
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
A little experiment: on my cluster I marked two OSDs from different hosts as "out".
After some time when remapping finished I stopped both OSDs and swapped them between hosts (with moving corresponding sections of ceph.conf and physical relocation of hard disks).
When OSDs started on their new hosts ~10% of cluster became remapping+backfilling but no data were flowing to or from those (empty) OSDs because they were still "out".
From common sense prospective rebalancing data among other 11 OSD was entirely unnecessary.
I hope this can be improved. Is it possible to make crush a little bit more intelligent?
Updated by Greg Farnum about 7 years ago
CRUSH improvements are a continuously ongoing discussion, and it's being improved right now.
Updated by Greg Farnum almost 7 years ago
- Project changed from Ceph to RADOS
- Category deleted (
10)
Actions