[Pulp-list] How about we just merge these core features into Cobbler?
Mairin Duffy
duffy at redhat.com
Fri Sep 12 18:30:48 UTC 2008
Michael DeHaan wrote:
> Máirín Duffy wrote:
>> Michael DeHaan wrote:
>> Backtracking a bit, I had one question. Is it a concern that if
the-artist-formerly-known-as-pulp becomes a part of cobbler, that it
would be difficult to tie in pulp functionality with a different
provisioning system? I had thought the original notion of having two
separate apps was partly to provide that kind of flexibility since
sometimes folks who manage their software distribution might not have
any control over the provisioning of machines or the software/process
used to provision the machines. Would pulp still be able to tie into
another provisioning system if it was built into cobbler?
>>
> Basically we have two classes of Cobbler users now:
>
> - ones who use the repo management bits
> - ones who don't
Well, to be fair my concern/focus isn't on Cobbler users, but on
RHEL/Fedora users trying to manage and deploy systems. I would guess
that the folks who are using cobbler in professional deployments today
are doing so because they have a great more deal of leverage over the
selection of management tools to use than the hypothetical users my
original concern was considering.
>
> I see the pulp features as being extensions on the existing repo
management bits, for the most part, though we'd probably want to discuss
them one by one on the lists.
>
> I am not sure everyone really wants a seperate app for each function
of things, more so, they just want tools that are easy to integrate
together.
>
> Using the cobbler repo management bits w/o the provisioning aspects
works today, so yes, it would not require that anyone use "cobbler
distro add" or "cobbler profile add" and similar features, just "cobbler
repo add"...
Okay that's good to know.
>> I think provisioning systems and managing the content to be
provisioned to those systems are separate tasks. I could see a benefit
to having a UI workflow that pulls together pieces of each integrated in
one UI, but managing a software distribution and provisioning systems
are still separate tasks and I think presenting the intricacies of both
all together in one UI might be a bit overwhelming.
>>
>
> It's blurred. Provisioning essentially means "giving out resources",
so not only can distributions be provisioned, but also packages, also
things like IP addresses and hostnames (which cobbler also does if so
configured).
Sure, that makes sense. I suspect, though, that there are a lot of
intricacies to cobbler-provisioning that are maybe too detailed or
in-depth or not commonly-used for the sort of workflow I was envisioning
that pulp could support. I just worry about the interface getting too
crowded and complex for what should be a simple workflow that just
happens to span a few different types of tasks, you know?
Maybe it would help to analogize what I mean: if I just want to make a
peanut butter sandwich, I don't need to be offered an apron, french
chopping knife, a 36-inch long cutting board, and a food processor to do
so. I mean, maybe the food processor would be handy if I wanted to shell
and crush the peanuts and make freshly made peanut butter for my
sandwich, but that's definitely a path less travelled :) But if I'm
Martha Stewart making a Thanksgiving feast, then at least some of those
tools probably are absolutely essential. Is the type of person who makes
PB&J and mac&cheese going to be able to do so with a Martha Stewart
kitchen (TM)? Yeh, but it might be a lot more confusing/complicated for
them ("which knife do i use?" "what does this tool do?") And it's not a
simple vs complex dichotomy, add a cafeteria chef into the mix who cooks
for 500 people a day, and he'll have a third separate workflow from
myself and Martha and maybe needs even more different tools.
>
>> Let me explain how I'm thinking this could work, based on some of
the stuff I've been working on My Fedora with J5, Eve, Luke, and Toshio.
Maybe it's not applicable, or maybe it is or would spark a good idea. As
you know, koji and bodhi are separate applications, geared for different
tasks (building packages and pushing updates) but those tasks are
related. Each is part of a larger 'package maintenance' workflow. (There
are other overarching workflows involving the two tools too, such as
release engineering but let's focus on pkg maintenance for now.) Our
plan for the My Fedora webui is to provide integration between the two
apps, koji and bodhi, in one UI tailored for a basic package maintenance
workflow. But the bodhi and koji UIs will still remain, they're not
going away, for more specialized tasks related to each respective
domain. Does that make sense?
>
> Perhaps. I guess a related question is, does anyone really like that
these are two seperate apps? I use bhodi for pushing updates, but never
really log into koji and just go by the email it sends me. If they were
better integrated where I could see the build logs when I was looking at
an update -- basically in the same app view, that might be easier.
Well, it's hard to say with My Fedora still not live and ready for use
yet (although we're getting closer!) From initial interviews and quick
paper prototyping with some of the folks here in Westford who use both
tools in package maintenance, I did get the feeling that there was a
desire to tie the two tools into one workflow, in a simplified manner,
but still allow for digging back into the original app (be it koji or
bodhi) in the exceptional cases where more detailed information/actions
needed to be taken. (Your example is one we handle this way - we tell
you in My Fedora if your build failed, and if you want to probe further
we link you to the appropriate spot in bodhi to pull up the build logs
to figure out what went wrong.) We'll learn more about whether or not
this suggested model works when My Fedora goes into production and
starts getting used.
>
>>
>> So I was thinking that maybe pulp could be a UI geared to the
current Satellite
build-your-distro-push-it-out-to-systems-and-update-your-distro-and-update-those-systems
workflow that Satellite (and Spacewalk :) ) users go through today. This
isn't to say there aren't other workflows that would use either/or or
both the repo management bits and cobbler, but pulp would be an
interface specifically geared towards the common Satellite/Spacewalk
workflow we know so well from working with and going out and
interviewing customers of the Satellite product.
>
> If it's geared to that workflow, why is it not part of Spacewalk? I
think that package management (RPMs) is the core use of Spacewalk today
... with the other features as being very useful but kind of a "bonus".
So maybe it could be just done as upgrades to that project if it's
more about that workflow?
Well, I do think there was some thought to it eventually replacing
Spacewalk. I am not sure if a 'clean slate' is needed there or not. If
cobbler is getting integrated into spacewalk, does that make spacewalk
the horse that ate the dog (cobbler) that ate the cat (pulp?) that ate
the mouse (my sanity?)
>
> Either way, cobbler's repo stuff could remain the backend. I am less
interested in what happens with the Web details (they are very
important, don't get me wrong), but am mainly interested in seeing we
leverage available bits on the backend.
Seems reasonable :)
>
> If Pulp would want to be a WebUI that supported the Cobbler API,
maybe that does make sense, but seeing the linkage that exists today
where we can associate profiles with repos in Cobbler, I like them being
together.
I'm not quite following this - you're saying you'd like pulp and cobbler
to be together, or you'd like the webui to be together with cobbler?
~m
More information about the Pulp-list
mailing list