Project

General

Profile

Feature #15880

crush:force creat pg

Added by ceph zte almost 8 years ago. Updated over 7 years ago.

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

0%

Source:
other
Tags:
Backport:
Reviewed:
Affected Versions:
Pull request ID:

Description

When ceph deply on some hosts.But the hosts are have diffrent osds.Some host have a lot and other have a few.
When we use the rule that diff host.And we want more data to the hosts that have a lot osds.So we change the
osd weight on the hosts that have few osds.Then after a long time the pg recover.For example we have 2000 PGs.
After the osd weight modefy most of the pg are active+clean. But there are a few pg such as three or four pgs that
are active+remapped.This is beacause that we change the osd weight.Fot change to active+remapped pgs to active+clean
we need to either modefy the osd weight back to 1 or change the cursh choose_local_fallback_trie.But the two menthod are
not we want.Can we add other methond.Since most of the pgs move from the lower osd host to the morer osd host.So we can
tolerate few pgs still on the lower osd host.
we can add force creat pgs in crush.To these fewer active+remapped pgs we can force creat them use the cli.The user should have the right to do that.After force creat pgs we can get a warn message in the ceph -s command.And use ceph health detail command we can known which pgs are force creat.After we add new host to cluster.Then we can cancel the force creat pgs.
Thank you!

History

#1 Updated by Sage Weil over 7 years ago

  • Status changed from New to Rejected

I think a better appraoch would be to set a pg_temp entry for the pgs so they'll show up as created but remapped. shoudl already be possible via 'ceph osd pg-temp <pgid> <osd1> <osd2> ...'

Also available in: Atom PDF