First, it starts with a DVD with an OS installer, then is a recipe to
automate it. And then PXE booting from network which sometimes lead to
discovering new bare-metal nodes. Some shops fully automate provisioning
and the only action is to power on a server. We all have that. But
what's next?
Make it fast. Blazing fast. Do it virtualization-like or cloud-like
experience with Foreman Discovery and the new image-deployment
capabilities.
Additional info:
Bare-metal remains relevant with upcoming farms of ARM servers. We need
to be ready to deploy dozens of bare-metal servers at once. I will show
story of provisioning in a typical Foreman shop and then make a demo of
bare-metal servers deploying OS using multicast image transfer.
The image-deployment scenario is a working in progress (prototype that
works), I would like to hear feedback on that and also collect ideas.
> I just submitted for the record:
>
> Abstract:
>
> First, it starts with a DVD with an OS installer, then is a recipe to
> automate it. And then PXE booting from network which sometimes lead to
> discovering new bare-metal nodes. Some shops fully automate provisioning
> and the only action is to power on a server. We all have that. But
> what's next?
>
> Make it fast. Blazing fast. Do it virtualization-like or cloud-like
> experience with Foreman Discovery and the new image-deployment
> capabilities.
>
> Additional info:
>
> Bare-metal remains relevant with upcoming farms of ARM servers. We need
> to be ready to deploy dozens of bare-metal servers at once. I will show
> story of provisioning in a typical Foreman shop and then make a demo of
> bare-metal servers deploying OS using multicast image transfer.
>
> The image-deployment scenario is a working in progress (prototype that
> works), I would like to hear feedback on that and also collect ideas.
Sounds pretty good to me. Just to check we're on the same page, the
patch below is meant so that the discovery image accepts blocks, so that
the discovery image can accept another image and boot from it?
> > I just submitted for the record:
> >
> > Abstract:
> >
> > First, it starts with a DVD with an OS installer, then is a recipe to
> > automate it. And then PXE booting from network which sometimes lead to
> > discovering new bare-metal nodes. Some shops fully automate provisioning
> > and the only action is to power on a server. We all have that. But
> > what's next?
> >
> > Make it fast. Blazing fast. Do it virtualization-like or cloud-like
> > experience with Foreman Discovery and the new image-deployment
> > capabilities.
> >
> > Additional info:
> >
> > Bare-metal remains relevant with upcoming farms of ARM servers. We need
> > to be ready to deploy dozens of bare-metal servers at once. I will show
> > story of provisioning in a typical Foreman shop and then make a demo of
> > bare-metal servers deploying OS using multicast image transfer.
> >
> > The image-deployment scenario is a working in progress (prototype that
> > works), I would like to hear feedback on that and also collect ideas.
>
> Sounds pretty good to me. Just to check we're on the same page, the
> patch below is meant so that the discovery image accepts blocks, so that
> the discovery image can accept another image and boot from it?
>
I was under the impression you could use image to dump to disk and boot it
(think reusing your vm templates / cloud images for bare metal too).
this can open the usage case of cloud init for bare metal and ideally speed
up bare metal provisioning time.
Ohad
···
On Tue, Sep 6, 2016 at 11:35 AM, Daniel Lobato Garcia wrote:
> On 09/05, Lukas Zapletal wrote:
> > Sounds pretty good to me. Just to check we're on the same page, the
> > patch below is meant so that the discovery image accepts blocks, so that
> > the discovery image can accept another image and boot from it?
>
> I was under the impression you could use image to dump to disk and boot it
> (think reusing your vm templates / cloud images for bare metal too).
>
> this can open the usage case of cloud init for bare metal and ideally speed
> up bare metal provisioning time.
Yes, today if you use the patch, you can dump the image using netcat or
using udpcast (multicast) and reboot and it just works. My idea is to:
extend FDI API with partitioning endpoint (create volumes/LVM)
create a image "library" plugin that will serve images
provide some helper scripts to feed the library with some images (*)
create new provisioning workflow in discovery plugin
Initial "library" could be very easy smart-proxy plugin with simple
provider (raw files in a directory) but we could integrate with other
technologies later on.
(*) virt-builder is a nice tool that can easily create bunch of images
for you with firstboot scripts, custom root passwords and stuff like
that
