MAAS fails to deploy Ubuntu on host that had CentOS deployed and vice-versa

[Lirim]

MAAS failing to deploy Ubuntu 16.04 on host that had CentOS 7 previously deployed with the message not enough space left on disk.After doing a manual install of Ubuntu,enlisting and commissioning the same node the deployment works fine.

[Andres Rodriguez (andreserl)]

Hi Lirim,

Unfortunately, we cannot help with this bug until you actually have some more information. Please provide:

1. MAAS version (dpkg -l | grep maas)
2. Curtin version (dpkg -l | grep curtin)
3. Curtin Installation log. (you can grab it from the UI).
4. Machine events (full list of events).
5. MAAS logs (/var/log/maas/*.log)
6. Cloud-init logs from the installation ephemeral environment.
7. rsyslogs (/var/log/maas/rsyslog/<machine-name>/<date>/messages)

[Andres Rodriguez (andreserl)]

After digging a little into this, this could be related to the recent UEFI fixes for CentOS that were made.

[Lou Peers (louie-pe)]

My understanding is that we discovered this behaviour last week, I'm not
sure what the timing of the updates were...

Lou Peers | Cloud Project Manager | Canonical
Google site: https://sites.google.com/a/canonical.com/customer-success-home/
irc: kip
DD: [phone]
M: [phone]
www.canonical.com | www.ubuntu.com

On 27/04/17 16:29, Andres Rodriguez wrote:
> After digging a little into this, this could be related to the recent
> UEFI fixes for CentOS that were made.
>
> ** Changed in: maas
> Assignee: (unassigned) => Blake Rouse (blake-rouse)
>

[Lou Peers (louie-pe)]

Hi guys,
Have we managed to find the source of the issue?
Thanks,
Lou

[Blake Rouse (blake-rouse)]

The issue is that curtin installs Ubuntu with '--no-nvram' this causes Ubuntu to not be installed as a bootloader in the EFI firmware. So when MAAS is down and netbooting fails Ubuntu will not boot. What needs to occur is that Ubuntu needs to be added to the EFI firmware but then the boot order needs to be changed so that the boot method used to go into the ephemeral is placed before Ubuntu. That will allow the machine to still be controlled by MAAS when commissioning or re-deployment needs to be done, but when MAAS is down it will boot from its local disk.

Here is the fix I did for the curtin-hooks for CentOS 7 that solves this problem for CentOS.

https://code.launchpad.net/~blake-rouse/maas-images/fix-centos-efi

[Blake Rouse (blake-rouse)]

This is related to bug 1680917. When this bug is fixed it will also fix the other bug.