adding a disk with "juju set" will chown every single file on every single osd

[Junien F]

Hi,

I you have a running ceph cluster, and you add disks via "juju set ceph-osd osd-devices", then for each added disk, the charm will chown -R /var/lib/ceph (https://github.com/openstack/charm-ceph-osd/blob/master/lib/ceph/utils.py#L1894)

However, /var/lib/ceph/osd has links to all osds, so the charm will chown every file on every osd. For every disk you're adding. So if you're adding 5 disks, the charm will chown -R 5 times, which takes quite some time and is not useful.

I'm not sure why we're chowning in the first place, but I feel like we should only chown what needs chowning. Filesystems which are already in ceph shouldn't need to be touched.

Thanks !

[Chris MacNaughton (chris.macnaughton)]

osdize_dir is only called with filesystem path OSDs; looking briefly at the code path at https://github.com/openstack/charm-ceph-osd/blob/master/lib/ceph/utils.py#L1420 , it looks like the charm _never_ calls chownr on a disk based OSD

[Junien F (axino)]

I guess I should have said I added *directories*, not disks, ie :

juju set ceph-osd-cache osd-devices="/srv/ceph/ceph0 /srv/ceph/ceph1 /srv/ceph/ceph2 /srv/ceph/ceph3 /srv/ceph/ceph4 /srv/ceph/ceph5 /srv/ceph/ceph6 /srv/ceph/ceph7 /srv/ceph/ceph8 /srv/ceph/ceph9"

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/623258

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/623258
Committed: https://git.openstack.org/cgit/openstack/charm-ceph-osd/commit/?id=87e43ce27a909ffc9d6516cd2f51a479a9b8a7e0
Submitter: Zuul
Branch: master

commit 87e43ce27a909ffc9d6516cd2f51a479a9b8a7e0
Author: Przemysław Hausman <email address hidden>
Date: Thu Dec 6 17:48:09 2018 +0100

    Remove chown'ing on adding dirs to osd-devices

    When adding directories to osd-devices (e.g. juju config ceph-osd
    osd-devices="/srv/ceph1 /srv/ceph2"), the charm was recursively changing
    owner and group for all files below /var/lib/ceph, for each added
    directory. This was taking time and was not useful.

    The fix is to remove unnecessary recursive chown on /var/lib/ceph, since
    ceph already manages the ownership of the files itself.

    Change-Id: I39a00591bc86ec49c4ced53eadce75ddb21e2431
    Closes-Bug: #1795383