Subversion migration - follow-up

[Ross Golder <ross golder org>]

It seems I stand accused of not announcing that Subversion was to be
hosted on a non-RedHat machine, and of making that decision without
consulting the board.

On examining the evidence - i.e. the archives of gnome-sysadmin ang
gnome-infrastructure - it does indeed seem I am guilty as charged. It
was indeed a terrible act of negligence and it does paint a pretty bad
picture of the way I handled the Subversion migration. My only defence
is that I had mentioned my intentions in the SubversionMigration wiki
page, which I've sent links to and encouraged people to read on many
occasions, but with hindsight it wasn't very prominently placed, not
worded very well at all and I can now see how some people might not have
realised my intentions. For some reason, I was under an assumption that
most people understood the situation and that nobody had any objections.

However, that is clearly not the case. It seems that I failed in my
responsibilities to the rest of the team, among other things. As such, I
hereby resign from the team so as not to be in a position to do any more
damage. Please remove my /etc/passwd entries, gnome-sysadmin list
membership etc etc. I've just started working full-time and am expecting
my first son to be born in a few days, so you should probably also
remove me from the accounts and moderator team memberships. I'll
continue to monitor gnome-infrastructure in case of questions.

Also, I should probably try to explain the rationale behind the
decision. Sometime after the failed migration in July I started working
on a new migration script (with more safety checks etc). I also had
communications with Michael Haggerty (cvs2svn hacker), who had
incorporated an improved version of my clockskew patch upstream, so I
started running tests using a more recent version of cvs2svn.
Unfortunately, the newer version of cvs2svn wouldn't work with the old
version of python running on container. At this point, I also learned
that subversion 1.2 which comes as standard on container (v1.2) was not
a good version for a large-scale subversion server for various reasons.
Also, as I was going to be running some pretty processor-intensive tests
on all 800+ modules and container is already pretty heavily loaded, I
really didn't want to slow CVS down for everyone any more than it
already was. At this point I was prepared to admit defeat and announce
the migration as cancelled/indefinitely postponed or whatever. Since
then, it appears that newer python packages have been made available for
RHEL3 and installed on container, but no further testing of the
migration scripts has been conducted there since.

Encouraged by messages from other GNOME hackers also eager to move on
from CVS, I decided to find a more appropriate platform to run the tests
and keep working on it. To begin with, I 'borrowed' progress.gnome.org
(the L10N pages server), with Danilo's consent and worked on there.
However, I also asked Canonical if they had a server (hardware
equivalent to or greater than container), and they provided
socket.gnome.org (around mid-November, IIRC). Socket has several
advantages over container, including better hardware a more up-to-date
base platform, and remote console/reboot access for available to the
GNOME sysadmin team among other things. Canonical also have support
staff on-site if required. I spent a considerable amount of time setting
up Socket, documenting it's setup on the wiki and ensuring, from a
sysadmin team perspective, it is as much a GNOME server as our other
servers.

However, that doesn't excuse the fact that I failed to announce more
formally through the proper channels before the migration the fact that
I planned to use Socket as the live subversion server going forward. It
was very unprofessional. I'm sorry to have let the team down again and I
hope my resignation from the team means that it will not happen again.

--
Ross