Project

General

Profile

Bug #474

mon: improve paxos commit batching

Added by Sage Weil almost 9 years ago. Updated almost 9 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
Category:
Monitor
Target version:
Start date:
10/10/2010
Due date:
% Done:

0%

Source:
Tags:
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:

Description

We should commit immediately if we haven't committed in the last 2 seconds. Currently we delay 2 seconds from the first change to batch changes, but that means single updates always take 2 seconds. This could even be a bit adaptive; the only reason for the delay is to avoid doing too many commits too quickly, generating lots of incrementals.

History

#1 Updated by Sage Weil almost 9 years ago

  • Assignee set to Greg Farnum

#2 Updated by Greg Farnum almost 9 years ago

  • Status changed from New to Resolved

Pushed a fix to unstable in 99013badb676986deb82757b77d91d0aa1f54cc9.
Instead of waiting g_conf.paxos_propose_interval seconds to comimt, we now wait until paxos_propose_interval seconds after the previous commit. If it's been more than paxos_propose_interval seconds, we wait g_conf.paxos_min_wait seconds. I set this to .05 seconds by default, in the hope that it will prevent a single OSD spawning multiple epochs on failure reports or something.

Also available in: Atom PDF