Project

General

Profile

Fix #13126

qa: ceph-fuse flushes very slowly in some workunits

Added by Greg Farnum over 8 years ago. Updated over 7 years ago.

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

0%

Source:
Development
Tags:
Backport:
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(FS):
ceph-fuse, osdc
Labels (FS):
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

Our ffsb and fsync workunits both generate lots of small IOs at random offsets, each of which becomes its own Op to the OSDs. That means they can take upwards of ten minutes to flush after the test "finishes". There might be other users with patterns like this, and it would be great if they didn't see such gratuitously IO-intensive bad performance.

History

#1 Updated by Greg Farnum over 8 years ago

This is blocked by/equivalent to #13127 in the Ceph project.

#2 Updated by Greg Farnum over 7 years ago

  • Category changed from 45 to Performance/Resource Usage
  • Status changed from New to Resolved
  • Component(FS) ceph-fuse, osdc added

I spot-checked one wow that the ObjectCacher is coalescing IOs to a single object. It looks like things have gotten faster but still aren't great because there are so many individual objects in a large file...which we probably can't do anything about? Closing since at least nobody's complained about it again.

Also available in: Atom PDF