Re: [libvirt] [PATCH 00/23] Introduce a virtlockd daemon for disk locking
On 08/09/2012 09:20 AM, Daniel P. Berrange wrote:
This is a long overdue update to a patch series I posted about
a year ago
https://www.redhat.com/archives/libvir-list/2011-July/msg00337.html
There have been some major changes since that series
Getting closer! I think the overall design is sound, and I have now
completed reviews of the entire series (instead of getting stuck at the
beginning); looking forward to a final rebase to clear up the questions
I raised on this round of review.
Still todo
- Add ability to quiesce all server/client I/O when doing
re-exec()
- Add ability to save/restore data in any virNetMessagePtr
structs in the client rx or tx queues
- Add ability to use custom lockspaces for LVM and SCSI/ISCSI
block devices, instead of locking based on path, to gain
cross-node safety, instead of node-local safety.
NB, the current re-exec() support works, but is not race safe
without those first 2 todo items being completed
Yeah, but it's still a strict improvement over the 'nop' manager for
out-of-the-box experience, so I don't mind if we push this series for
0.10.0 even without those to-do items resolved yet. I think the design
is extensible enough that we aren't locking ourselves into an inability
to implement them.
--
Eric Blake eblake(a)redhat.com +1-919-301-3266
Libvirt virtualization library http://libvirt.org
Attachments:
- signature.asc (application/pgp-signature — 620 bytes)