I hate to lead with this sort of subject. This is actually a serious
question and *not* flame fodder :)

Based on the volatility of the Debian squeeze release, I'm switching
to lenny. I'm now bumping into the XFS kernel panic cataloged here:

http://developer.amazonwebservices.com/connect/message.jspa?messageID...

There's apparently a way to side step this using a log level when
formatting XFS. That's cool, but not a happy feeling to be one command
option away from a kernel panic :(

But I'm wondering though if Ubuntu isn't a better, long term distro
for EC2, at least for my requirements. Specifically, I'm looking for:

- A very lean, minimal server config
- Stability, but *reasonable* currency (I've found many of the lenny
packages to be shockingly behind: e.g. python-boto version 1.2??, bzr
version 1.5???)
- "Path of least resistance" with EC2 operability

I don't mind working through issues, but I feel apprehensive about
Debian at the moment. Of course, my requirements aren't representative
and most people I'm sure are perfectly happy with Debian on EC2.

Debian has historically been known for its conservatism, so that
seemed to me the path to take. But is there perhaps a particular
Ubuntu release that's  closer to what I'm looking for?

Garrett

Garrett:

To clarify what you are asking about, the issue here is not Debian vs.
Ubuntu, but "Debian on EC2" vs. "Ubuntu on EC2" which are completely
different beasts.

Historically on EC2, neither Debian nor Ubuntu have had completely
Debian and Ubuntu stacks from the kernel up.  Instead, we've had to
make do with patching the runtime on top of older kernels (2.6.16,
2.6.18, 2.6.21) built for other distros (Fedora Core).

Amazingly, we've managed to hold this together and keep it running
well for a couple years on EC2.

As of a few weeks ago, however, we no longer have a kernel which is
both secure and which supports XFS (without the workaround you allude
to).  Additionally, the newer Debian (squeeze) and Ubuntu (Karmic)
releases have a version of udev which will no longer function on a
2.6.21 kernel.

I've been working with the Canonical server team for a while, and in
recent months they have put a lot of effort into making solid, secure,
consistent, Ubuntu 9.10 Karmic AMIs for EC2 along with modern matching
kernels and ramdisks, using an automated and reproducible process.
Release candidates are available today and the official release is in
5 days.

Canonical has stated that their next focus will be to apply the same
process to releasing quality Ubuntu 8.04 Hardy AMIs with kernels and
ramdisks.  Hardy is older, but it is an LTS (long term support)
release which a lot of folks are already using on EC2 with the older
Amazon kernels.

I'm not trying to convince anybody to switch from Debian to Ubuntu,
but just wanted to point out that what is coming out for Ubuntu on EC2
is going to be real Ubuntu, while Debian on EC2 does not currently
have the same development efforts being applied and is going to be
exposed to greater risks over time.

My focus the last couple years on EC2 has been to build solid Ubuntu
AMIs which I can employ for my personal and business use.  I have
published them publicly because I wanted to share the results of my
efforts, and in return I have also gained tremendously from the
contributions of the community.

I have also been publishing Debian AMIs because it wasn't too much
additional work once I got Ubuntu working.  I've been happy to hear
that these AMIs have been useful to many developers.

The plan for Ubuntu on EC2 is to migrate folks over to using the new
images from Canonical and/or build their own Ubuntu images using the
newer kernels from Canonical.

I've stated that I would continue publishing Ubuntu AMIs as long as it
makes sense.  With a funded company officially publishing quality
Ubuntu AMIs and expanding the versions they are supporting, I think
we're getting close to the point where it won't be useful for me to
publish base Ubuntu server images.

When the point comes where I am not publishing the base Ubuntu server
AMIs, it will not make sense for me to publish new Debian AMIs
either.  At this point, I'm not even able to work on solving some of
the issues that Debian is facing, like the fact that squeeze doesn't
work with Amazon's 2.6.21 kernel.

One advancement that might help Debian on EC2 is the new Ubuntu
kernels and ramdisks being built by Canonical.  I haven't tested them
with Debian myself, but I would encourage Debian users on EC2 to start
trying to figure out if this can be made to work.  I could include
patches in ec2ubuntu-build-ami that improve building Debian AMIs for
the community.

I'm happy to chat further with official or unofficial Debian
developers about the future of Debian on EC2.  I'm anticipating that I
will be withdrawing from this project in coming months and would like
to help make a smooth transition to one or more other projects which
could continue building and publishing Debian AMIs on EC2.

I've already published the code which is used to build the AMIs so
this is effectively an open source project ready for branching.
Another approach would be to enhance the "vmbuilder" project which
Canonical is using to build their Ubuntu AMIs.

The initiative here has to come from the Debian community.  I look
forward to seeing what this evolves into.

I'll reiterate that I will not be deleting the Debian AMIs which are
already published.  They will continue to be available for the
foreseeable future.

--
Eric Hammond

On Oct 24, 6:35 pm, Garrett Smith <g...@rre.tt> wrote:

Yes, yes, yes!

I very much appreciate this. The build script has been very well
written and maintained.

-snip-

This is helpful in navigating this issue. Fortunately Ubuntu and
Debian are consistent enough to make a switch fairly easily --
certainly more viable than going over to the RPM line.

-snip-

With the strength of the Debian community, I'd be surprised if this
wasn't taken up by capable hands. Thanks to your efforts and those of
other contributors, there's a solid base to build on.

Garrett

Hi Garret and Eric,

I've built one Debian AMI using the Canonical's kernel version 2.6.28
and it works smoothly.
Unfortunately, I still don't have a feedback from the Amazon EC2
community on how this AMI works, but I do know by my own experience
that it works fine till now.
Regarding the Canonical's kernels, the only issue I've noticed while
using their kernel in this AMI is the support for RAID, that is not
currently present in the kernel.
I'm planning to officially support the development of Debian AMIs, as,
just like Eric, I've been using for my personal and business use.
Eric, if possible for you, I'd like to spread this information within
the Debian and Ubuntu groups and to have your own evaluation of the
AMI would be the most appreciated.
The AMI id is ami-2d759644.
All input from the community will also be pretty much welcome.
Garret, perhaps this AMI fits in your requirements.

--
Ramses Paiva

On Oct 25, 2:18 pm, Garrett Smith <g...@rre.tt> wrote:

Ramses:

Glad to see your initiative with Debian on EC2.

More important than the AMI itself is having clear documentation
describing how you built the AMI.  This lets others duplicate your
efforts securely and with their own customizations, not worrying about
whether your  AWS account will be around as long as they need to use
the AMI.  And, folks who trust you can use your AMI with more
understanding about what went into it.

Think of it as an open source project and be as open and welcoming as
you can.  This is a lot of extra work, but it helps the entire
community and you benefit in the end as others contribute to the
project or ask for your expert consulting assistance.

--
Eric Hammond

On Dec 10, 11:33 pm, ramspaiva <rpa...@debianzone.org> wrote: