Project

General

Profile

Bug #42509

ceph balancer should consider number of OSD available for the pool

Added by Марк Коренберг over 4 years ago. Updated about 4 years ago.

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

0%

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

Description

For small pools (i.e. cephfs metadata for example) it considers too small number of pools. For example, for my 16-OSD cluster, it suggests 4 PGs. If I agreed with that, I would lose IOPS since data will not be distributed by all of my OSDs.

So, balancer should limit the minimum number of PGs to at least, say, 3*<number of OSDsavailable for the pool>.


Related issues

Copied to mgr - Backport #42999: nautilus: ceph balancer should consider number of OSD available for the pool Resolved

History

#1 Updated by Greg Farnum over 4 years ago

  • Project changed from Ceph to mgr

#2 Updated by Sage Weil over 4 years ago

  • Status changed from New to Fix Under Review
  • Pull request ID set to 31636

#3 Updated by Sage Weil over 4 years ago

  • Backport set to nautilus

#4 Updated by Марк Коренберг over 4 years ago

no, simple increase of the minimum number of PG is not what I wanted.

#5 Updated by Sage Weil over 4 years ago

  • Status changed from Fix Under Review to Pending Backport

#6 Updated by Nathan Cutler over 4 years ago

  • Copied to Backport #42999: nautilus: ceph balancer should consider number of OSD available for the pool added

#7 Updated by Марк Коренберг over 4 years ago

So, why are you ignore my messages about considering current number of OSDs ? But instead you have hardcoded constant of 16 ?

#8 Updated by Nathan Cutler about 4 years ago

  • Status changed from Pending Backport to Resolved

While running with --resolve-parent, the script "backport-create-issue" noticed that all backports of this issue are in status "Resolved" or "Rejected".

#9 Updated by Марк Коренберг about 4 years ago

I still disagree. Bug not solved completely.

Also available in: Atom PDF