Hi Darryl On Wed, 3 Feb 2016 at 00:39 Darryl Weaver <dar...@darrylweaver.com> wrote:
> This would therefore have implications for co-locating nova-compute with > ceph charms on physical nodes. > Upgrading nova-compute would update the apt repository regardless of what > the ceph charm is doing. > Yes - this is one of the drawbacks of pushing two services onto the same underlying machine. Fortunately in this case, only the ceph binaries on disk are updated by the package upgrade - the package maintainer scripts are explicitly configured to not restart ceph services, so that this can have appropriate rolling restart orchestration. > Presumably ceph would then upgrade the next time config-changed hook runs? > I was thinking more along the lines of: juju set ceph source=cloud:trusty-liberty # no-op reconfiguration of software sources juju action do ceph/0 upgrade # update && dist-upgrade && restart of services on unit juju action do ceph/1 upgrade juju action do ceph/2 upgrade This is esp. important for moving to >= infernalis where the ceph daemons run as the ceph user instead of root, requiring a filesystem permissions update for the ceph osd filesystems (which could be very slow depending on storage capacity of the cloud). Just as a sidenote; we're doing some work between now and 16.04 that will mean we deprecate the ceph charm in favour of a new 'ceph-mon' charm - this just runs the monitor process for a ceph cluster and can be placed in a LXC container, allowing for better control over physical server resources for the container once Juju makes the move to LXD. So deployments will be ceph-mon + ceph-osd - the storage team are working on a migration approach for existing ceph deployments as part of this plan.
-- Juju mailing list Juju@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/juju