Re: [Xen-devel] Developmentstatus for Xen with Ceph as replicated storage

[George Dunlap <dunlapg@xxxxxxxxx>]

On Thu, May 24, 2018 at 12:01 PM, thg <nospam@xxxxxxxxx> wrote:
> Hi everybody,
>
> in 2013 there was an announcment, that XenServer will fully support the
> RBDs from Ceph, to use them as blockdevice for VMs (see
> <https://ceph.com/geen-categorie/xenserver-support-for-rbd/>)
>
> What is about Xen itself, how it is supported? I know that you can map
> an RBD as device and use it for putting a VM-image on it. But this is a
> "manual" process and thus not usable for cloud-servers with many VMs.
>
> Anybody who has experiences with this or an other (working) option?

Well fundamentally *something* has to convert block read/writes into
network packets on the Ceph protocol.  There are three basic ways this
could be done:
1. The guest could have a Ceph driver, and speak to Ceph directly.
2. You could have a user-level process which speaks both Ceph and some
other protocol (say, the Xen PV protocol or an emulated disk) that
does the conversion; for example, QEMU.
3. You could have the dom0 kernel do it.

#3 is what you describe here.

The issue I suspect you're facing is that you don't want to have to
manually create device nodes every time you create a guest, and copy
that device node in to the config file.  The solution to that for
upstream xen is block hotplug scripts.  You can see examples in
/etc/xen/scripts/block-*; for example, block-iscsi.

It looks like someone has started to do that work here:

https://github.com/FlorianHeigl/xen-ceph-rbd

The warning there about it not working with pygrub is out of date --
that should work now (but you'll probably want to give it some good
testing just in case).

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel