[Biopython-dev] Updating http://biopython.open-bio.org/SRC/biopython/

[Bartek Wilczynski]

On Tue, Nov 17, 2009 at 1:16 PM, Peter <biopython at maubp.freeserve.co.uk> wrote:
> Dear all,
>
> Back when Biopython used CVS, we had an hourly checkout of
> the code published here:
> http://www.biopython.org/SRC/biopython/
> http://biopython.open-bio.org/SRC/biopython/
>
> I did ask how this was setup (which username etc) but no one got
> back to me. In any case, this can now be turned off - along with
> making the Biopython CVS read only (OBF support call #857).
>
> I've managed to get this working again from the github repository,
> although the route isn't ideal (and is all running under my username):

Great!
>
> (1) At quarter past the hour, a cron job on dev.open-bio.org does
> a "git pull" to update a local copy of the repository to that on github.
> This doesn't need any passwords.
>
> (2) At 25mins past the hour, a cron job on www.biopython.org does
> an rsync to get a copy of the repository from dev.open-bio.org (using
> an SSH key for access).
>
> I tried having a single job running on dev.open-bio.org to push the
> files to www.biopython.org but the host policies seem to block that.

that's a pity. In case you haven't thought about it, I would only
suggest to use some sort of a lockfile: In case github is very slow
(happens every so often to me) it might occur that the second job will
start before the first one is finished. Leading potentially to a
broken repository for download. If the first job would create a
lockfile before it starts and would delete it after it's done, the
second job could condition the rsync operation on the existence of the
file. This way, we could have delaysin syncing , rather than
potentially broken repo. Of course, if we move to a simpler setup
using only one host, this would not be needed.

>
> It would be simpler to have everything done on www.biopython.org,
> which would require git to be installed on that machine. This would
> avoid any SSH security problems. Does that seem like a better idea?
>

indeed, this would be best, and it only requires someone with root
privileges to install a package.

Also, it would be cool if the biopython user was somehow unlocked.
(Currently no-one seems to have the password...)

I don't hav an account on www.biopython.org server, so I can't help
much at this point...


cheers
Bartek