De : Daniel P. Berrange [mailto:berrange@redhat.com] Envoyé : vendredi 8 novembre 2013 06:32 À : Eric Blake Cc : NEVEU Stephane; libvirt-users@redhat.com Objet : Re: [libvirt-users] RBD images locking On Thu, Nov 07, 2013 at 09:08:58AM -0700, Eric Blake wrote:
On 11/07/2013 09:04 AM, NEVEU Stephane wrote:
Eric,
[please don't top-post on technical lists]
Well, in case where several servers may start the same virtual machines after a reboot for exemple. http://lists.ceph.com/pipermail/ceph-users-ceph.com/2013-August/0038 87.html
Isn't the existing virtlockd support already sufficient for this? If not, what is preventing the virtlock framework from interacting with rbd disks?
Nothing really. The current impl only deals with disks with type=file or type=block, ignoring type=network. We could extend it to cope with the latter though - either using a URI to uniquely identify the >network storage, or better yet using some unique volume ID if available.
Yes it might be useful when not using an orchestrator. Can't we use the "block_name_prefix" of a volume to do so ? It seems to be unique and to be the entry point to identify an image (via objects) across a pool : rbd -p mypool info myvm rbd image 'myvm' Size 10240 MB in 2560 objects Order 22 (4096 kB objects) block_name_prefix: rb.0.11a6.238e1f29 format: 1 rados -p mypool ls | grep ^rb.0.11a6.238e1f29 | wc -l 630 objects