Project

General

Profile

Bug #58606

osd: osd_recovery_cost with mClockScheduler enabled doesn't reflect the size of the object being recovered leading to degraded recovery performance..

Added by Sridhar Seshasayee about 2 months ago. Updated about 2 months ago.

Status:
Fix Under Review
Priority:
Normal
Assignee:
Sridhar Seshasayee
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:
49975
Crash signature (v1):
Crash signature (v2):

Description

Currently osd_recovery_cost is set to a static value equivalent to 20 MiB.
This cost is set regardless of the size of the object being recovered.
This results in the recovery items, regardless of object size, to remain
in the mClock queue for the same amount of time. This negatively impacts
the performance of recovery ops.

To resolve this, osd_recovery_cost must reflect the size of the object
being recovered before enqueuing the item into the mClock queue.

See related BZ for more details: https://bugzilla.redhat.com/show_bug.cgi?id=2163473

History

#1 Updated by Sridhar Seshasayee about 2 months ago

  • Subject changed from osd: osd_recovery_cost with mClockScheduler enabled must reflect the size of the object being recovered. to osd: osd_recovery_cost with mClockScheduler enabled doesn't reflect the size of the object being recovered leading to degraded recovery performance..

#2 Updated by Sridhar Seshasayee about 2 months ago

  • Status changed from New to Fix Under Review
  • Pull request ID set to 49975

Also available in: Atom PDF