setup.py install and bdist_egg

[Hyuga]

On Mar 13, 4:41 pm, Jasiu <jasi... at gmail.com> wrote:
> Hey,
>
> I work at a company where I'm lucky enough to write web apps using
> Python and WSGI :). We develop more and more stuff in Python and it's
> becoming a mess of dependencies, so we thought we would create a
> guideline for developers that describes the whole process of deploying
> a Python app on a server. Since all of our servers run Debian, all of
> our software has to be installed as Debian packages - other
> departments that don't use Python (shame on them! :D) already follow
> this policy.
>
> Here is what we have figured so far:
> 1) Write your code.
> 2) Prepare setup.py, and put names of required eggs in there.
> 3) Write a Debian wrapper that uses CDBS Python class, and put names
> of required Debian packages in there.
> 4) Build Debian package.
> 5) Done! Deploy on server and have fun :).
>
> This guideline already works pretty well for us, but I thought I could
> tweak it a little. I want to make the Debian package wrapper as thin
> as possible so that if we ever stop using Debian, we will still be
> able to deploy all of our software as eggs. This goal turned to be
> pretty challenging. I have a few questions.

Glad to see I'm not the only one with this sort of challenge.  Though
it sounds like you're already a step ahead of me in that you're
properly maintaining requirements/dependencies for your Python
packages.  When I build my Debian packages I make sure that *they*
have the correct dependencies.  But my setup.py files do not have all
their dependencies listed, so it's a pain to get everything set up
correctly when not doing it through the Debian package system.

> 1) Debian has a debian/dirs file where you can list all the
> directories that the package should create, for example those in /var/
> log, /var/lib . I already figured out that I can use setup.py
> 'install_data' keyword for that, even for empty directories. Then I
> can use 'setup.py install' command and things work great. But... How
> about eggs? I'd like my egg to somehow contain or create such
> directories. For example, I have a config file that I want to place
> in /etc directory. I can do that using 'setup.py install', but
> 'setup.py bdist_egg' will just create an egg containing etc/
> directory. Can I do something about it? Also, why doesn't an egg
> contain empty dirs that I place in install_data?

My approach to this has been to use a postinst script to copy files
out of the eggs (or when I'm not installing as an egg, from /usr/
share) to the correct locations in /etc.  However, this doesn't work
so well if you're not using a Debian package.  But see below:

> 2) What about file permissions? Can they be set by setup.py? My
> software runs as www-data user and so /var/log and /var/lib
> directories should have proper owner, group and permissions. I didn't
> figure out a way to change permissions using setup.py. In Debian I can
> use postinst script for that.

I'm also using a postinst script for configuring my system, but I want
to move away from that.  I already have a command-line tool for easily
administering certain aspects of the system.  So my plan is to
implement a "deploy" command that creates all the directories with the
correct permissions, generates the config files, and performs all the
configuration my postinst script currently does.  Then the postinst
script simply needs to call the deploy command.  That way I can take
advantage of debconf without being dependent on it.

> If anything sounds unclear, blame my bad english :).

Sounds perfectly clear to me!