When is it better to just use containers?
On 1/28/23 18:43, Demi Marie Obenour wrote:
On 1/28/23 18:06, Nico Kadel-Garcia wrote:
> On Thu, Jan 26, 2023 at 8:31 PM Reon Beon via devel
> <devel(a)lists.fedoraproject.org> wrote:
>>
>> Are there still some outstanding bugs preventing this from happening?
>
> Is there any one critical feature that justifies the update? Avoiding
> the requirement of python is... OK, maybe understandable, but I don't
> see it as a "must-have" improvement. And better modularity support....
> My observation so far is that modularity simply destabilizes systems,
> because the authors of the "modularized" tools do not build up the
> full suites of likely necessary components. I'm running into that
> right now with python310 back in RHEL 8 for ansible, the results are
> not pretty.
At this point it might be better to just containerize Ansible.
Containers may waste resources and require extra effort to keep
up-to-date, but they *work*, and that is important. They also
*massively* reduce the test burden.
To elaborate: if I am an upstream developer for something like Ansible,
my options are either:
1. Ship a whole bunch of packages for a whole bunch of distributions
and make sure everything works across all of the various
dependency versions. This also means that I have to restrict
myself to the packages that e.g. RHEL 8 has, which might be
quite old.
2. Ship a single container that only needs to be QA’d once, works
everywhere, and has no dependencies except for the Linux kernel.
I get to update dependencies when *I* want to, and don’t have
to worry about breaking user’s systems. I can even ship a
shell script wrapper so that the container can be invoked as
a binary.
3. Something that I am not aware of (suggestions welcome).
It’s pretty easy to see why someone would go with option 2.
--
Sincerely,
Demi Marie Obenour (she/her/hers)