Project

General

Profile

Bug #39330

recovery transfer rate not correct

Added by Jonas Jelten 12 months ago.

Status:
New
Priority:
Low
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:

Description

When running all OSDs inside a QEMU VM (with real disks attached through virtio-scsi), the gathered recovery statistics rate does not equal the real IO rate (measured on the hypervisor):

ceph shows:
659 active+clean
109 active+remapped+backfilling
recovery: 1.6 GiB/s, 403 objects/s

This reported IO rate does not match the measured IO rates of QEMU (iotop or htop):

qemu disk read: 236.00 MB/s
write: 218.72 MB/s

All the PGs being backfilled due to rebalancing are 8+3 erasure coding.

Any idea where the calculation is wrong and the result ends up being way too much?

Also available in: Atom PDF