git + fabric = awesome deploy team
I just rewrote my web site Wacchen in python with Django. I converted it from a rails project. It’s easy to say something like rails sucks, but clearly people have been very successful while using rails, so I am not going to say that. I am going to say that I love Django, and the process of building my project from the ground up in Django was fun. But, rails ain’t all bad, they do have a cousin called Capistrano.
Capistrano isn’t actively developed anymore, but it still kicks ass. It helps you build a one line deployment script, which is really-really important. For years, I would start to build a script like that, and then at some point I was like, fuck it, I can do it later. What I didn’t realize was that I was throwing one of the best getting shit done tools out of the window. Once you taste one line deploys, a lot of mental blockage fades away, and you can iterate faster. I had never realize how powerful the thought of trying to deploy was. It was actually stoping me from want to hack. Now though, I converted to python land, and I needed something like Capistrano. Django doesn’t have as tight a relationship with Fabric as Rails does with Capistrano, but it’s been mentioned in more then one place.
That is what I chose. I started with the above, and an assortment of
blogs posts, but fabric has changed a lot since many of these posts were
written. For instance, it’s not config anymore it’s env. Also you can’t
use the fancy text replacer syntax any longer. This
run('cd $(path); mv releases/current releases/_previous;') now becomes
this run(‘cd %(path)s; mv releases/current releases/_previous;’ %
env)`.
I am glad to say that it works, and it’s been quite plesant. I have learned a couple things along the way that I can share though.
There is no package deal
Borrow, cheat, look around, but you aren’t going to be able to use someone elses script verbatim, deployment is personal. This was a huge stumbling block for my self in the beginning, but like I said capistrano was like crack, so I got myself over the hump.
Use git
The ability to have git just dump a tar of your current directory is
super useful. Then all you need to do is take that tarball and put it
on the server. This was the extent of my putting.
This get’s me using git too, which is a nice added bonus. I use git, but I often forget at the very beginning to use it.
Use your fabfile as a scratch pad
Because there is no cannon for fab files, and you can’t just use someone else’s I found it best to create new deployment steps as I went. I would ssh to the server, mess with the command until it was right, and then immediately write it into my fabfile.
Chunk it up
Don’t try and write big fab targets, use lot’s of little ones, that way as your deployments get longer, you can start to target smaller portions of the deployment. I have an app server, and celery server, and I didn’t want to deploy too both servers all the time.
Those are my quick notes for now. I also ended up using a lot of other new tools like gunicorn, nginx, dreamhost ps, celery, and rabbitMQ, all of which have their own set of notes.