With MAAS 1.9a2, lvm disks are not correctly skipped when in-use by ceph disk prepare.

[Mark Shuttleworth]

Today's gmaas autopilot run seemed to stumble on Ceph. MAAS 1.9b2 might be an issue.

ceph-disk-prepare skips disks that are already mounted, however it misses a new case where maas 1.9 now uses lvm by default.

One approach would be to make sure the disk is not a physical volume in an active volume group with mounted logical volumes as another conditional to check.

[David Britton (dpb)]

Indeed.

http://paste.ubuntu.com/13102405/

unit-ceph-1[20040]: 2015-11-04 08:46:46 INFO unit.ceph/1.mon-relation-changed logger.go:40 ceph-disk: Error: Device /dev/sda1 is in use by a device-mapper mapping (dm-crypt?): dm-0

[David Britton (dpb)]

I believe by default, maas 1.9 now uses LVM, which probably means that this check in ceph/charmhelpers is no longer sufficient (as the device is no longer mounted, the LV ends up getting mounted):

    if not os.path.exists(dev):
        log('Path {} does not exist - bailing'.format(dev))
        return

    if not is_block_device(dev):
        log('Path {} is not a block device - bailing'.format(dev))
        return

    if (is_osd_disk(dev) and not reformat_osd):
        log('Looks like {} is already an OSD, skipping.'.format(dev))
        return

    if is_device_mounted(dev):
        log('Looks like {} is in use, skipping.'.format(dev))
        return

[David Britton (dpb)]

FYI to others who hit this, I believe there is a global switch in maas 1.9 to change to 'flat layout' that should workaround the issue.

[Mark Shuttleworth (sabdfl)]

MAAS 1.9 will do LVM and bcache and RAID, and the Autopilot does need to
support those *well*, not just document a workaround.

Mark

[Chris Holcombe (xfactor973)]

I have a WIP branch for this that I've linked. I'm still giving some thought to how to properly test it

[Wade Fitzpatrick (wade-fitzpatrick-p)]

My servers have the following disk configuration:

/dev/sda - 240GB SSD
/dev/sdb - 1TB SATA
/dev/sdc - 1TB SATA
/dev/sdd - 1TB SATA
/dev/sde - 1TB SATA

What I want to do is use LVM to partition 40GB of the SSD for a root volume then use the rest of it for Ceph but it doesn't look like the installer will allow that because it seems to want whole disks.

I worked around this bug by partitioning the SSD with a /boot on /dev/sda1 which gets mounted so the checks in the ceph installer skip over that disk. How can I tell it to use the other 200GB of the SSD for ceph though?

[Chris Holcombe (xfactor973)]

@wade-fitzpatrick-p sorry for the delayed response. The installer will allow you to specify a directory as your OSD so what you could do is mount your 200GB SSD to some mount point and then tell the charm to consume that mount point.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/308295
Committed: https://git.openstack.org/cgit/openstack/charm-ceph/commit/?id=57a305f40e422e99897d59d75e3f28e47d24feb6
Submitter: Jenkins
Branch: master

commit 57a305f40e422e99897d59d75e3f28e47d24feb6
Author: Chris MacNaughton <email address hidden>
Date: Wed Apr 20 08:12:00 2016 -0400

    Charmhelpers Sync, fixes mount detection

    This charmhelpers change corrects how we detect
     if a device is mounted

    Closes-Bug: #1513009
    Closes-Bug: #1571840

    Change-Id: I9310e7aae1957d6c44e6a3e48c4426e34960a63a

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/308296
Committed: https://git.openstack.org/cgit/openstack/charm-ceph-osd/commit/?id=5d30ecf502effbae2e35363fbd89e04a3cadf51c
Submitter: Jenkins
Branch: master

commit 5d30ecf502effbae2e35363fbd89e04a3cadf51c
Author: Chris MacNaughton <email address hidden>
Date: Wed Apr 20 08:12:27 2016 -0400

    Charmhelpers Sync, fixes mount detection

    This charmhelpers change corrects how we detect
     if a device is mounted

    Closes-Bug: #1513009
    Closes-Bug: #1571840

    Change-Id: I5c42fba38d11dd1cc02723dbe8e1b6bae7d5d35d