[Openstack-operators] Ephemeral instances in RBD issue

Abel Lopez alopgeek at gmail.com
Wed Oct 29 19:34:36 UTC 2014


So, I figured out why this was happening.
The gist of it is, the direct_snapshot doesn't start the domain back up. I contacted inktank and they're adding what I found to their fork.

The long story:
in nova/virt/libvirt/driver.py, the snapshot method does everything. the code doesn't support live snapshot for lvm or rbd, so we have to do a cold snapshot.
We go into 'managedSave' ( the instance is suspended to quiesce I/O), the snapshot is taken, then much later, we start a new domain based on the save.
The last commit to jdurin's fork splits snapshot right after the 'try direct snapshot' and adds a 'generic_snapshot' method. The code for starting up the domain again is in 'generic_snapshot' which only gets called if the exception "ImageUnacceptable" gets raised.
as a quick hack, I just copied the 'new_dom' lines into 'snapshot' method and HUZZAH, it works as expected.

Hopefully we'll see a new commit to jdurgin's fork for icehouse and havana soon.

On Oct 26, 2014, at 6:08 AM, Simon Leinen <simon.leinen at switch.ch> wrote:

> Abel Lopez writes:
>> I saw this last commit to jdurgin's nova fork which solves the issue (
>> https://github.com/jdurgin/nova/commit/ea4b5369e4bec4dd7a0ce9f68769600329cda6c6
>> )
>> now a snapshot happens in seconds. 
> 
>> The problem that we've introduced however, is that about 15-20m after
>> we do a snapshot, the VM is powered off.
>> Every time.
> 
> Ouch! Have you checked the logs (nova-compute and maybe libvirtd's)?
> 
>> I can start the instance back up with `nova start`, but I am leery of
>> pushing this out to prod and having to tell users to expect a shutdown
>> after a snapshot.
> 
> Understood.
> 
>> Anyone else using this in Havana?
> 
> Not me, but I'm sympathetic with your worries, and want this resolved as
> well.  We're using Icehouse with RBD, currently without the "ephemeral"
> patches, but we would really like to (re-) activate that part of the
> integration soon.
> 
> It's maybe worth asking on #ceph or posting to one of the CEPH mailing
> lists, too.
> 
> Good luck,
> -- 
> Simon.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 496 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20141029/d55caf5f/attachment.pgp>


More information about the OpenStack-operators mailing list