Mailing List Archive

Hi Torin

On Mon, 12 Mar 2018 at 21:52, Torin Woltjer <torin.woltjer@granddial.com>
wrote:

> Hello,
>
> I am looking to deploy an openstack cluster using LXD for compute and Ceph
> for storage, and I was running into some doubt as to whether this was
> possible; and doubt that nova-lxd was mature enough for production. If
> anyone is running nova-lxd in production, or knows anything about it,
> please let me know. I've had a hard time finding good informational
> resources on the topic, specifically relating to LXD + Ceph; which is
> supposedly possible, but I haven't heard if it's possible in Openstack. If
> you otherwise know of a resource that could be helpful to me, I would
> appreciate hearing it.
>


Short answer is that nova-lxd does support use with Ceph, but only for
additional block devices (I.e no boot from volume or ephemeral device
support right now).

You have highlighted a documentation gap - there are a few non obvious
things todo like ensuring that cinder creates RBD devices with a minimal
feature set to support use the the kRBD driver used for nova-lxd. Will
look to put a howto for ceph in place shortly!

Cheers

James

Thank you for the response James.

I now have a couple of further questions regarding boot volume support on nova-lxd.

Is this feature on the radar?
On nova-kvm documentation states you need shared storage for live migration; is this the same case with nova-lxd, or can you live migrate between compute hosts when using a dir storage pool for root?
Putting the host's LXD storage under a folder that a ceph pool is mounted on, while an obvious sleight of hand, what would the repercussions be?

I don't know if anyone has answers to these, but any are welcome. I'm assuming the feature I'm looking for relys on work from the nova project rather than the LXD project; I will try to track down a nova features timeline or submit a request myself.
James, any documentation you can put together would be great and I look forward to seeing it.

Thanks.

----------------------------------------
From: James Page <james.page@ubuntu.com>
Sent: 3/13/18 5:33 AM
To: torin.woltjer@granddial.com
Cc: openstack@lists.openstack.org
Subject: Re: [Openstack] Nova + LXD + Ceph?
Hi Torin

On Mon, 12 Mar 2018 at 21:52, Torin Woltjer <torin.woltjer@granddial.com> wrote:
Hello,

I am looking to deploy an openstack cluster using LXD for compute and Ceph for storage, and I was running into some doubt as to whether this was possible; and doubt that nova-lxd was mature enough for production. If anyone is running nova-lxd in production, or knows anything about it, please let me know. I've had a hard time finding good informational resources on the topic, specifically relating to LXD + Ceph; which is supposedly possible, but I haven't heard if it's possible in Openstack. If you otherwise know of a resource that could be helpful to me, I would appreciate hearing it.

Short answer is that nova-lxd does support use with Ceph, but only for additional block devices (I.e no boot from volume or ephemeral device support right now).

You have highlighted a documentation gap - there are a few non obvious things todo like ensuring that cinder creates RBD devices with a minimal feature set to support use the the kRBD driver used for nova-lxd. Will look to put a howto for ceph in place shortly!

Cheers

James

Hi Torin

On Tue, 13 Mar 2018 at 13:59 Torin Woltjer <torin.woltjer@granddial.com>
wrote:

> Thank you for the response James.
>
> I now have a couple of further questions regarding boot volume support on
> nova-lxd.
>
> Is this feature on the radar?
>

Not right now; I'm not entirely sure its even possible with LXD but I've
not dug into this in huge detail to date.


> On nova-kvm documentation states you need shared storage for live
> migration; is this the same case with nova-lxd, or can you live migrate
> between compute hosts when using a dir storage pool for root?
>

Its possible to block migrate as part of the live-migration process for
libvirt/kvm; With regards to live migration in LXD - this is definitely an
experimental feature right now; CRIU is improving all of the time and
live-migration might just work for your workload but there are lots of
common cases that probably don't.


> Putting the host's LXD storage under a folder that a ceph pool is mounted
> on, while an obvious sleight of hand, what would the repercussions be?
>

I think this is probably OK right now, but its not something I've
personally tried as most of our validation for nova-lxd is based around
2.0.x series, which does not have support for storage pools + ceph.


>
> I don't know if anyone has answers to these, but any are welcome. I'm
> assuming the feature I'm looking for relys on work from the nova project
> rather than the LXD project; I will try to track down a nova features
> timeline or submit a request myself.
>

nova-lxd is still a separate team and project:

https://github.com/openstack/nova-lxd

so its a little different to main Nova development right now; a bug report
at:

https://bugs.launchpad.net/nova-lxd

is a good place to start!

Cheers

James