On the last day of DebConf 13, RichiH and Ganneff and I set up a git-annex repository that all Debian developers have automatic commit access to, and everyone can clone and browse. See the announcement email.

This is an interesting repository for me for several reasons. Selfishly, I have always wanted an easy way to download lots of DebConf photos for offline viewing, and never seem to get around to downloading everything from gallery.debconf.org when I have bandwidth. I've also wanted an example repository that shows how git-annex can be used by a large group for collaboration. Finally, the way this repository is set up with an incoming queue is fairly unique.

With 430 files in the repository, totaling over 3.5 gigabytes (which doesn't include all the talk videos that are #included into it), and at least 18 people having cloned the repository so far, the debconf-share repository is well on its way to being a sort of large git-annex repository.

Just running git annex whereis is interesting; many of the files already have 8 copies. Some talk videos are more popular than others and you can see when they're downloaded too. But enough snooping.. ;)

So far people have uploaded mostly photos and talk slides. Other places exist to store those things in the DebConf infrastructure, but it's nice to have them all available in one tree. I particularly like today's addition of chrysn's files which include the raw photos and hugin files used to produce panoramas, and then pull those together into a postcard which has all its sources available.

In my corner of the debconf-share repository, I'm collecting together files regarding the possibly-historic dpkg-source-git-re-re-redesign process that would have otherwise been scattered around various places and probably not all published. This includes an hour long recording of the main design session (recorded with permissions) made by my laptop's mic, which, surprisingly, turned out to be pretty listenable. I will probably have more to say about this process later, once Ian announces dgit.

So, we're still seeing how usage develops. I hope that having this available during the next DebConf, and other Debian meetings, rather than only at the end, will further facilitate file sharing and storage. Especially if a fast clone is available right on the DebConf LAN. ;)


The technical details of how the repository is put together are:

  • There's a repository on git.debian.org, which piggy-backs on the collab-maint group, so most Debian people have commit access to it.
  • git-annex is used to upload files to that repository, as an incoming queue.
  • A git-annex-shell annex-content hook is run whenever someone uploads a file to there. It moves all annexed content over to annex.debconf.org for publication. This involves some ugly but safe stuff to do with publically readable restricted use ssh private keys. It was the hardest piece to get working, and is only necessary because we don't want to bloat git.debian.org with this stuff and it's not practical to give everyone logins to annex.debconf.org.
  • As an additional guard against accidental bloat, the git.debian.org repository will refuse to accept uploads when there is less than 5 gb free disk.
  • annex.debconf.org was set up following the git-annex public repository on a web site howto.

I've found at least 4 bugs in git-annex as a result of this repository, which is a rather unusual use case. And fixed 2 of them so far..

copies?
Can git-annex ensure that, even if a pile of random developers have a copy of a file, the copy in the shared repository is still required?
Comment by Josh
comment 3
The repository can be cloned anonymously by: git clone http://annex.debconf.org/debconf-share/.git/
Comment by joeyh.name
comment 4
@Josh the way this setup ensures the copy is the central repository is always required is that the central repository is only accessed via http, so read-only. This prevents someone dropping files from it by accident. It would be possible for DDs to do so maliciously, with rather a lot of setup work -- and lots of us have backups if someone did that..
Comment by joeyh.name