Re: [atomic-devel] EC2 PV/HVM AMIs

[Colin Walters <walters verbum org>]

On Mon, Jul 21, 2014, at 09:56 AM, Andy Grimm wrote:
>
> The only  issue with making a paravirtualized atomic image is that we
> don't currently generate the required grub.conf file as part of our
> image build.  It would be trivial to do so (the Fedora cloud-base
> kickstart has code for it that could be reused); it just wasn't a
> priority.

There's a few more issues here.  One is that EC2 (even HVM images AIUI)
have a special "hvc0" console.  The current cloud image has that in the
kernel args for grub1, but not for syslinux.  That way the same binary
image works both in EC2 and say OpenStack/oVirt/bare libvirt.

Unfortunately that trick won't work for Atomic, at least for updates,
because ostree doesn't know about GRUB 1 config files.
 
> This is techinically correct, but perhaps I can make it more clear.
> Joe's response was that project atomic would not "own" any images
> produced.  Images *will* be produced by Fedora, and probably CentOS as
> well (and some of the same people will probably be involved), but it
> makes sense for the distros to own the image production process and
> the resulting artifacts.

Yeah, this gets to the second aspect, which is the current images are
generated by rpm-ostree-toolbox create-vm-disk, but that code was a
stopgap until we had Anaconda.  Now that's landed, the question is how
to use it.

That's where ImageFactory comes in - it wraps Anaconda in a VM.  And so
I think that's the next step for generating the Atomic image - try
writing a Kickstart to pull from a tree compose.

I just started sketching out some work here:
https://git.fedorahosted.org/cgit/spin-kickstarts.git/commit/?id=b73614d1e19f75e189b9f90e90e596b19d3b9314
I'm traveling at the moment but will try running ImageFactory when I get
home.