Re: Debian needs more buildds. It has offers. They aren't being accepted.
On Mon, Feb 09, 2004 at 04:26:32PM +0100, Stefano Zacchiroli wrote:
> Could you or Ingo please post to this list a brief summary of those
> "(plenty of) information"? Or even a pointer to mailing list archives
> or whatever ...
Well, first let me make clear that this is not going against the person of James
T. or Ryan M., I really do appreciate their work and involvement as such,
but I get the impression that they're too loaded with work and can't keep up
with it as well as their seem to be some pesonal objections by those against
others that result in very unnice situations.
For example last autumn there was a big backlog for the mips port. I offered
two (well, first only one, but there were actually two) MIPS machines as a
buildd. They already were configured, running and online and just needed
wanna-build access. And there even was an experienced buildd admin to
maintain them. But the offer was rejected for some strange reasons. Here are
some of them:
- "you're incompetent and have no knowledge about MIPS port"
- "you're no DD and there's more than having a machine to be able to run a
buildd"
- "it's the best to have just one buildd admin for that work to minimize
overhead"
- "there's a new big machine coming soon to handle that backlog, no need for
your slow machines"
- ...
(served by memory)
Obviously most of the reasons are plain stupid.
I was never asked about my knowledge about MIPS. In fact I worked longer on
SGI machines than I'm using Debian. In one of my prior jobs the company was
a registered SGI Developer. I'm running a m68k buildd for years now,
although just as the host admin, not the buildd admin. That part is done by
the same DD who was intended to admin the buildd on the mips machines.
The argument of having just one buildd admin is arguable as well. Sure,
having multiple buildd admins does require a basic will to communicate and
cooperate. The m68k port is an excellent example of how well this can be
done and how much a port can benefit from multiple buildd admins. You may
want to have a look at debian-mips ML archive and search for buildd related
mails.
Although there might be a nice, fast and big mips buildd in the make, it's
not online yet and not working. My two offered machiness might be slower,
but they were available and usuable at that time I offered them. There were
more offers for buildds on debian-mips as well.
For mipsel there was a huge backlog caused by not building qt-x11-free for
some weeks, although it was on top of the needs-build queue.
Some people were nearly complaining daily on debian-mips about that and
finally someone built it privately and uploaded it. Apparently, the buildd
admin was totally unresponsive.
That's about mips.
For m68k the situation is as this:
There have always been complaints about m68k being a doorstopper arch and
being dead slow. Sure, it's not as fast as a new 3 GHz Intel, but it is a
very nice arch and has still many friends and users.
So I organized some more machines for the m68k port and there is still
Goswin Brederlow with two "fast" 060s to help out. He's activily helping in
compiling and debugging for the m68k port. Therefore there was the idea t
let his machines run as buildds as well. That was rejected because "he is no
DD" and it was said, that it would be acceptable when a DD was taking the
buildd part and sign the packages. So it happened. Wouter Verhelst mailed
the needed info to James and Ryan (I think), but for more than two weeks
nothing happened and no info available on that as well.
In the meantime Stephen Marenka setup a new buildd (jt7) and got w-b access
within a day or so. We then setup akire shortly after that and waited two
weeks for getting w-b access for it. Ok, buildd.d.o moved to newraff in that
time as well and Wouter seemed to have made a mistake in mailing, but James
(elmo) was unresponsive to us (Goswin, Wouter and me) on IRC as well,
although he was actively ircing from time to time.
Well, long story, short result:
We were forced to actively take and handle packages by hand to bring the
backlog down, which is really a lot of work. That's why buildd was invented,
wasn't it? ;)
We were put back in our effort by having slow (not to say non-existing)
communication with either James or Ryan. We have some more buildds in the
make, some of them could be used as public machines for all DDs as well, and
we have to work on debian-installer for m68k as well, but there we have some
problems.
When we shutdown some buildds, we can work on d-i but generate a huge
backlog. When we work on bringing the backlog down, we can't work on d-i.
That's why I have tried to bring another 3-4 060 buildds online over the
last weeks, running against the "getting w-b access"-wall.
That's rather frustrating and annoying.
I personally have two conlusions about this whole issue:
a) If you work fine with Goswin and cooperates with him nicely, you'll run
against walls when you ned to interact wih James or Ryan. I do know that
some people have strong objections against Goswin for years now. But I think
as well that personal dislikes shouldn't interfere with some role position
work as it is the position of being a DAM or giving w-b access to others,
that might result in negative side-effects for a whole arch.
b) The flow of communication is very bad. I understand that working for
Debian is a voluntary work, but so it is for me. As well I understand that
other people has some "serious" work as well. So do I have. And therefore I
need information by those on which my work for Debian rely. That way I can
plan and schedule my own work.
For example the move of buildd and w-b o newraff.
Instead of informing the porters in advance that it will move and when, it
has just moved, giving some "problems" for the porters and many questions,
because packages uploaded seemed to got lost somewhere.
Or regarding w-b access for new buildds: I understand that moving buildd/w-b
to newraff was an important move, but I wished that we would have received a
short mail like "I got your request, but we're moving stuff to a new
machines right now and will add your machines to access when the move
finished next weekend". Nothing more needed and that would be a very good
thing because we would knew what's happening and that nothing is lost and
could concentrate on our things then.
So, when f.e. James is too busy with Debian work to be able to give small
explanations or infos by mail, a way should be find to broaden the burdon of
his work to more people. It's not a good way to have both sides left
frustrated and annoyed: the asking side and the asked side, which is flooded
with questions and such.
I think a good communication is essential for role positions as
documentation is for programmers and admins in a company.
As an admin you can't just move or shutdown a service in your company
without prior announcement - at least when you plan to stay longer employed,
that is... ;) (yeah, I know that Debian is not a company, but that doesn't
mean that we can't learn good things how to act like a ccompany, does it?)
Just another bad example of bad communication:
When Ryan announced the movement to newraff, I replied to him and asked him
to have a bzip2'ed version of the arch specific w-b info files (*-all.txt)
to save traffic (1.1 MB vs 11 MB each hour) for www.buildd.net. No response.
I'm already don't try to ask Ryan and James to support the automated status
update for their buildds, because I don't think that they would either
respond nor participate. :-((
Well, that's it, reported by memory and surely my own impressions that I got
over the last weeks and months. YMMV.
--
Ciao... //
Ingo \X/
Reply to: