On Thu, Oct 29, 2015 at 9:16 AM, Joe Brockmeier <jzb(a)redhat.com> wrote:
On 10/28/2015 08:21 PM, Josh Boyer wrote:
> The *could* be the same thing,
> except cloud-init is terrible and I hate it and if that was the single
> offering we had for some kind of C&S WG I would cry. I hate it
> because it is ridiculous to use in a non-cloud environment, and Server
> very much has that as part of it's reach.
Forking this thread briefly because I think this deserves its own
discussion.
I apologize if my rambling wasn't clear on this point. Hopefully this
tangent is short-lived.
Is your objection primarily to the concept of cloud-init or the
implementation? If it's the concept, not much we can help with there. If
it's the implementation...
Well, neither really. Admittedly my use of the Cloud images, and
therefore cloud-init, was in attempted to boot it in a VM and log in
more like a traditional install for simple test purposes. That didn't
work and getting it to the point where I could log in required running
some virt-tool thing to modify the image offline. So in the context
of "Server & Cloud", where people expect to be able to log in after an
install in many cases, cloud-init makes it really hard and is
ill-suited to that kind of environment.
Specific to cloud environments, I have no idea if the hassle of
getting it setup is the norm or worthwhile. I've been told it is, and
I can see where having the infrastructure setup to provide the
credentials already in place might make the hassle much less
problematic.
(It is also quite possible I hit a bug in the cloud image. I tried
running the local setup to provide cloud-init with ssh keys and it
didn't work, hence the virt-tool thing. It has been a while since I
tried again.)
We've talked about replacing cloud-init a few times in the past,
but
there are two objections:
- cloud-init is "standard" and we have an uphill marketing battle to get
our image adopted with something else.
- lack of a great alternative.
I completely believe both of these.
Mike has talked about a "rich boot process" previously, and
I wonder if
we're ready to start working on that?
I'm not sure what "rich boot process" means. I'd immediately
interpret that as "a real init process" which to me means using
systemd. Somehow I don't think that's what you're thinking... :)
Also, one of the CentOS GSoC projects was "Flamingo"
"a lightweight
contextualization tool that aims to handle initialization of cloud
instances." [1] Maybe this is something we could look at for F24? CC'ing
Tamer Tas, the student who worked on that. (It's targeted at being a
cloud-init replacement for Atomic, so...)
[1]
https://github.com/tmrts/flamingo
That might be nice for "get rid of python" reasons. If it had
cloud-init compatibility that would be even better, since people
wouldn't need to migrate their provisioning infrastructure.
josh