Sam Hartman (hartmans) wrote,
Sam Hartman
hartmans

The Case for Proposal B

This is my personal opinion, not that of the project leader. Tomorrow,
I'll write an essay trying to discuss the various options in with as
little bias as I can manage (although even that will be Sam's opinion).
Several people have asked me why I included Proposal B.
This is my answer.


While I was talking to people about systemd and init systems, people
seemed to inherently assume that being uncomfortable with systemd meant
that you were in favor of sysvinit, or at least init-script based
solutions. At least, people who were heavily involved in the issue made
that assumption. That didn't resonate with me.


Several concerns commonly raised with systemd resonate with me:


  1. It combines a bunch of things in one project; as an example how you
    start daemons ends up being tied to how you configure the network.

  2. This combination seems like it might reduce innovation at least
    outside of the systemd ecosystem, because interfaces are coupled.

  3. It is Linux specific


Of these, the biggest concern for me is the idea that systemd might
stifle innovation by becoming one point of control.


And yet, in my opinion, systemd is vastly superior to the current
alternatives. I'd far rather be writing service units than init
scripts. They are more declarative. Dependencies that I care about are
easier to express. There are better security isolation facilities. In
non-Debian work I've found that I depend heavily on systemd because it
is easier and more pleasurable to code to than the alternatives.
Declarative syntax for managing users is useful. I haven't personally
seen the huge joy of socket activation, but if I were writing somewhat
different things, perhaps I would. Given
the options today, I would pick systemd hands down and not look back.


But what about tomorrow? For me, one of the great things about Debian
has been that it's possible to integrate new technologies and to try
things out. Debian has been the OS where I and many others could try
out new technologies and figure out what it was like to fully integrate
them into the operating system. Systemd is the best we've got now, but
I'm reluctant to step away from Debian as a platform for innovation and
experimentation.


Yet I don't think focusing on sysvinit or other init-script based
solutions actually has anything to do with the kind of innovation I'm
talking about. I understand that for people who value sysvinit (or
something like runit) above systemd, that work is valuable. My
experience is that for my needs, systemd is a better fit. I wanted a
proposal that allowed us to maintain Debian as a platform for innovation
without focusing on the legacy of init scripts. I think that if there
is going to be something that some day replaces systemd, it will support
service units (or a significant subset) not init scripts. I suspect it
will have a way to handle socket activation and so on. And I cannot
imagine a future systemd replacement that does not have advanced
security isolation features.


How it Works


Proposal B is a systemd focused proposal. It's very similar to Proposal F.
The text is different, but the implications of both proposals are
similar. Maintainers can use whatever systemd facilities they choose.
Init scripts are not required. For most maintainers, even thinking
about alternate init systems or future experiments is something entirely
optional. That's true under both proposal F and Proposal B.


Where they differ is in how much support the project gives to
experiments involving alternate init systems. Under Proposal F, that's
entirely optional at each member's discretion. My experience is that's
not sufficient for Debian to remain a community for innovation. My
experience is that key maintainers and teams maintaining central
infrastructure or packages often need to work with people who are trying
to integrate new features. The difference between Proposal B and F is
that under Proposal B, we commit to making that happen for technologies
that are important in exploring alternatives to systemd.


Obviously, no member of our community is obligated to do work. In
practice this commitment might mean working to find new volunteers to
help out key packages or teams and do this work. Sadly, there are areas
where the history of interaction has not been so good; behavior on
multiple sides of discussions has not lived up to our standards. In
addition to making sure we have the necessary volunteers for key
packages and teams,
part of meeting this commitment may involve working with people who
want to explore alternatives to systemd to find advocates who foster a
climate where we can be excellent to each other.


The Risks


There are some real risks with Proposal B. The biggest is that we'll
spend time working on integrations and nothing innovative will come out
of it. A possible outcome is that we spend a lot of time integrating
elogind and similar technologies, but they end up not being useful
because packages start depending on service units and socket
activations. Unless something new comes along, we may waste our
effort. Yet we've often been willing to spend effort to enable people
to try things. For me, this proposal is about reaffirming that aspect
of Debian.


In the worst case, it's possible that we decrease the quality of our
systemd integration leaving room for something else, spend significant
emotional energy, and do not end up with interesting innovation.
I think it's much more likely that if there is no interesting
innovation, Proposal B will slowly morph into Proposal F.


Why did You Do this?


In the beginning of this post, I talked about how I personally
considered the concerns about systemd separate than the desire to keep
init-script based systems running. That view is uncommon among people
who have been spending a lot of time on this issue. In general people
who are spending a lot of time on init systems seem to be fairly
divided. If you are trying to get work done today, you are probably
either fully using systemd or using one of the existing init-script
based alternatives.


However, my concern resonated with developers I talk to who spend less
time involved in the issue. Not people who were going to go research
things enough to write a proposal. But people who weren't sure that
systemd was the way and the light of the future, but found it had a lot
of great things going for it.


I was one of the few people who was taking the time to really understand
the issues but who was somewhat flexible. I didn't even know how I was
going to rank the options on my ballot until this morning. Yes, I've
been systemd leaning in some ways, but I also very much see the
arguments in favor of enabling people to keep other init systems
working. I'd be happy with most of the options on this ballot winning.
So, I tried to listen and see if there were ways of splitting
disagreement that wouldn't work for the people most committed to one
position, but might appeal to people who are less involved.


Why are you Writing This Post?


I think it's dangerous for someone who is project leader to speak a
personal opinion, especially on a controversial issue. However, I've
heard people struggling with some of the issues I discuss here in our
community. What I say may give people another way of looking at
things. I do think I have a valuable prospective because I have spent a
lot of time thinking about the issues but have not been as intimately
involved as others who have spent similar time. I think my need to act
as a facilitator at least for this GR is over. And after spending a day
considering, I think it's more beneficial to specifically ask the
project to think about Debian as a community for experimentation than to
say nothing.

Tags: debian
Subscribe

Recent Posts from This Journal

  • Voting Guide for Debian Init Systems GR

    There are a lot of options on the ballot for the Init Systems GR. There have been hundreds of messages on debian-vote, and more scattered across…

  • Free as in Sausage Making: Inside the Debian Project

    Recently, we’ve been having some discussion around the use of non-free software and services in doing our Debian work. In judging consensus…

  • AH/DAM/DPL Meet Up

    All the members of the Antiharassment team met with the Debian Account Managers and the DPL in that other Cambridge— the one with proper behaviour,…

  • Post a new comment

    Error

    default userpic

    Your reply will be screened

    Your IP address will be recorded 

    When you submit the form an invisible reCAPTCHA check will be performed.
    You must follow the Privacy Policy and Google Terms of use.
  • 0 comments