On Wed, Aug 26, 2015, at 05:09 PM, Josh Boyer wrote:
Hi All,
I'm emailing my questions on the topic here as it seems to be the best
Fedora focused place to discuss Atomic Host and kernel interaction.
If that isn't the case, please point me to where you believe that is.
As jbrooks pointed out, there is also atomic-devel(a)projectatomic.io.
Regarding interaction, one thing I'd ask the Fedora kernel packaging
team to keep in mind whenever making any changes to the kernel.spec
is that ostree takes over kernel installation - we don't use the stuff in %post.
The atomic swap of the bootloader configuration is at the heart of the model.
Things work OK now, but just something to keep in mind.
I have two basic questions around the interaction of Atomic Host and
the kernel. The first is fairly straightforward: is there anything
Atomic Host or the atomic toolset needs that the kernel does not
provide today?
So the ostree side is one thing, but beyond that for Docker, storage
and networking are big items. I think conversations on storage for
containers are going on elsewhere, but just noting that.
The second question is a bit more involved. Atomic provides the
nice
ability for rollback across the entire OS tree. However, that
requires an atomic image to be spun for every instance of that tree.
I'd say:
"requires an new tree composed to be spun for updated kernels".
I use "image" for qcow2, installer ISO, PXE-to-Live etc. The first
two support in place upgrades.
That, naturally, means that whenever a new Atomic Host instance is
spun it will use whatever kernel happens to be the latest in the
Fedora release it is built from. This means that one cannot leverage
the nice side effect of being able to update the kernel independently
of userspace. (Which is also nice from a testing perspective when it
comes to kernels and regressions.)
Yes, currently, but most limitations around that kind of thing go
away with
https://github.com/projectatomic/rpm-ostree/pull/107/commits
To my understanding, the only way to provide such testing would be
to
create Atomic Host images that only deviate from the official images
in that they provide a new kernel.
There's also support for updates-testing:
https://fedoraproject.org/wiki/QA:Updates_Testing#Using_it_with_Fedora_At...
With a two week image release timeframe though, being able to use
different kernels might be a good idea. Does anyone have any thoughts
around this topic and how to possibly accomplish such testing? The
only other idea I had was to spin the Atomic Host images containing
the last 3 kernels in them, but I am not sure if choosing between them
at boot is currently possible with multiple kernels installed.
Having multiple trees that differ only by kernel by default in a single image
is indeed possible. It might be interesting to have updates-testing cloud
images generated that also have the non-updates-testing content or something
like that.
Finally of course, don't forget that composing trees is something one can
do locally too, it doesn't require magical rel-eng powers. It is an investment
to set up, but for those with multiple machines where you want to synchronize
state, it could be worth the investment.