[PVE-User] cloud-init problems/comments/questions

Wolfgang Bumiller w.bumiller at proxmox.com
Fri May 18 10:11:40 CEST 2018

Thanks for the detailed report, comments inline:

On Thu, May 17, 2018 at 02:50:12PM +0200, Klaus Darilion wrote:
> Hi!
> Thanks for cloud-support. I just tested it as you descibe on
> https://pve.proxmox.com/wiki/Cloud-Init_Support
> The only difference is, that I am using a zfs pool storage instead of
> local-lvm.
> Basically it worked out of the box, except:
> a) On boot, the Ubuntu cloud image reports:
> --- start console output ---
> startin serial terminal on interface serial0
> Booting from Hard Disk...
> error: no such device: root.
> Press any key to continue
> --- end console output ---

This one is an issue with the ubuntu image. There's an upstream bug
entry here: https://bugs.launchpad.net/cloud-images/+bug/1726476

> After ~ 15 seconds the boot continues, even without pressing a key.
> Maybe the VM creating is not 100% perfect.
> b) It would be great if you can add MTU configuration to the cloud-init
> network device. When using lots of OVS-tunnels, MTU must be lowered in
> the VMs.

Makes sense, can you open a feature request on our bugzilla please?

> c) I use ZFS replication for my VMs. The first issue encountered was,
> that replication for the linked clone did not worked. Even if
> replication for the template is set up und working fine. Is this
> intended to fail?

Not intended AFAIK. We're looking into it.

> Next I tried using a full clone. Also, the cloudinit disk was created on
> the zfs storage. Replication works now fine, but only zfs:vm-125-disk-1
> is replicated, the cloudinit disk zfs:vm-125-cloudinit is not
> replicated. Hence, migration fails with:
> can't migrate local disk 'zfs:vm-125-cloudinit': local cdrom image
> So, it seems currently cloud-init VMs with ZFS storage can not migrated.
> correct?
> Is this because it is not solveable at all, or just not done in this
> release. I think it should be possible, when replication also replicates
> the cloudinit-disk.

We'll have to include it in migration. I think initially there was the
idea to recreate it, but then there's no guarantee it's exactly the same
(different tool versions, timestamps etc.) so we need to add this.
Can you open a bug report?

> d) adding SSH keys: it is a bit confusing as the wiki uses "--sshkey"
> but the manpage defines "--sskeys". It should be identical, to stop
> confusing beginners like me.

Thanks, yes we should update the wiki.
The CLI tool should, however, accept both since it will autocomplete
shortened options on its own.

> e) adding SSH keys: it fails if the pubkeyfile contains lines with
> comments (#)

Admittedly I haven't thought of that (as ssh keys already contain a
comment part at the end anyway ;-))

> f) how can I add multiple nameservers? "--nameserver,"
> does not work. It is not documented how to add multiple servers.

Space separated. We should make this consistent with containers where
spaces, commas and semicolons can be used.
Thanks for noticing.

> g) "man qm" mentions:        --name <string>
>            Set a name for the VM. Only used on the configuration web
> interface.
> This is not true anymore - name is also used for the cloud-init
> hostname. The man page should be updated.

True, thank you. The cloud-init patch set had a couple of versions and
used to also have a separate hostname property which we removed to stay
consistent with containers later on and apparently forgot to update

> h) The VNC console behaves strange. The first time it always works. The
> second time, it takes ~15 seconds - maybe related to a) - to get output.
> Also, starting a second console at the same time, the second console
> does not work. This works fine with normal created VMs.

This is due to the way serial sockets are handled by qemu. There may be
some options to improve this but I'm not sure.
Also note that serial sockets are quite limited, eg. we won't be able to
let automatically redraw the screen when connecting, or have the guest
know about a change in size.

> i) when i change a cloud-init config in the webgui, and then click on
> "regenerate image", an Error is generated: "unable to parse directory
> volume name 'cloudinit' (500).
> After that, the vm-123-cloudinit.qcow2 file dissappeared from the
> storage. This seems like a bug.

Can you post the VM config and type of storage you use for the image? I
can't reproduce this here right now, it may also explain this post
here: https://forum.proxmox.com/threads/cloud-init-docs.43798/

More information about the pve-user mailing list