Merge lp:~hopem/charms/precise/cinder/lp1226823 into lp:charms/raring/cinder
Proposed by
Edward Hope-Morley
Status: | Merged | ||||
---|---|---|---|---|---|
Merged at revision: | 22 | ||||
Proposed branch: | lp:~hopem/charms/precise/cinder/lp1226823 | ||||
Merge into: | lp:charms/raring/cinder | ||||
Diff against target: |
51 lines (+23/-2) 3 files modified
config.yaml (+10/-0) hooks/cinder-hooks (+12/-1) revision (+1/-1) |
||||
To merge this branch: | bzr merge lp:~hopem/charms/precise/cinder/lp1226823 | ||||
Related bugs: |
|
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
James Page | Pending | ||
Adam Gandelman | Pending | ||
Review via email: mp+187289@code.launchpad.net |
This proposal supersedes a proposal from 2013-09-17.
Description of the change
* Sets minimum recommended pg settings when creating cinder rbd pool.
* Sets replication count of 3 when creating new cinder pool.
Fixes: bug 1226823
To post a comment you must log in.
Edward and I chatted about this on IRC and decided the minimal fix for the bug here is to set the PG number on the pool, and not the replication count. Hard-coding the replication count to 3 on the new pool is dependent on the ceph cluster on the other end. We'd still like to support deploying single-node ceph clusters for testing, and adding this to the cinder charm would block that. As far as I can tell reading docs, a PG number of 1024 should be supported on any size ceph cluster.
We need a better way of managing these details. I don't believe the remote client services should be dictating the particulars of the ceph pools they use. I'd be happier if, via the relation, a service like cinder could simply request the creation of some pools, eg ['cinder', 'images', 'otherstuff'] and it is up to the ceph charm to create those pools with parameters that are sensible for the size/capacity of the deployed ceph cluster, probably based on its charm config.
That said, if it safe to use a default 1024 PG count for all ceph clusters, I would be okay with having cinder set this itself when creating its pool. But I think we both agree the replication count should stay unspecified until we can come up with a better way of setting that.
I'm in the process of testing this change.