On Mon, Feb 1, 2010 at 10:11 PM, Greg DeKoenigsberg <gdk(a)redhat.com> wrote:
1. Fedora AKI for EC2.
Where we are: Chatted with Justin Forbes today briefly, and evidently
we're waiting for the 2.6.32 kernel to move from updates-testing to
release. The goal of the kernel work is to make sure that it actually
runs properly on Xen hosts, which is what Amazon runs EC2 on.
How to help: Amazon runs EC2 on many versions of RHEL. The oldest is
analogous to RHEL 5.2. If you can build a (RH)EL 5.2 Xen host, and
install a Fedora 12 guest built with the 2.6.32 kernel in updates-testing.
Justin, any particular things to test on that guest? Is "does the damned
thing boot" an effective test? Any particular set of tests?
Some amount of stress testing is probably worthwhile; some of the
kernels I've seen have fallen over under load even though they booted
2. Compose tools for Fedora EC2 images.
To some degree, this is parallel work. To complete it, we'll need a good
F12 kernel -- but to get it started, we just need an overview of the
current tools, figure out how they're used, document, improve if need be,
and make it dead-easy for Fedora users to do what they want to do. I
think a lot of people have ideas of what we *could* do, but we've yet to
articulate what we *will* do.
My $0.02: seems like thincrust might be a good place to start. Red Hat
folks have been working on it for a while, and I've liked my initial
experiences with it. It's dead easy to go from kickstart file to virtual
image, and they've also got a tool called ec2-converter that allegedly
sets up AMI images that are ready for upload. Maybe the place to start is
to take a current spin and walk down the path of turning it into an AMI,
and see how far we get. (Copying Joey Boggs, who is the maintainer of
ec2-converter: any advice? Gotchas?)
thincrust is still using the python-imgcreate stuff, correct? The
other gotcha is around kernel/ramdisk setting, but Joey called that
out. There are probably also a few other things around actually
setting the image up to be a little bit more usable with building up
some utilities to do some of the fetching of info off of the local
http node that gives you instance info. But also probably parallel.
3. IRC meetings.
This "roadmap", such as it is, is very high-level. I wonder if perhaps we
should convene an IRC meeting this week and figure out who's willing to
jump in and help. At the very least, we can articulate some tasks at a
more granular level, and maybe get some folks to pick them up.
So... um... Thursday at 1900GMT / 1400 Eastern US time? I've got
#fedora-cloud registered on freenode. We could hang out, put together a
roadmap... you know, junk like that.
I might not be able to make it during that time as it's smack dab in
the middle of one of the few meetings I can't miss. But I wouldn't
say that's a counter-argument. Friday afternoon is empty for me as a
counter proposal if it works for people.