Experimenting with fabric for deployments

Fabric has caught my eye as a deployment tool several times over the last year or so. Each time I’ve seen it, my thought has been that it’s a attempt to replace something that doesn’t need replacing, namely the venerable unix shell script.

Fabric came up again this week in an article by the NPR application team ¹ and on this occasion I saw that they were solving a similar problem to one that I recently solved with a shell script. The authors seemed clear-headed and not prone to fads so I thought I’d take the opportunity to duplicate my shell script functionality in a fabric fabfile.

I must say that I was impressed. The fabric tutorial is very well structured and allowed me to get results quickly. Within a few hours I’d git-rm’ed my shell script and the build-version_control-test-deploy cycle for this site is now handled by a fabfile. While the fabfile is the same length as the shell script, I think it’s far easier to read and allows partial deployments (reminding me of make). I particularly like the innovative use of context managers as it removes one of the key problems with deployment scripts, that inadvertently modifying global state can cause unexpected downstream side-effects. Well done, Fabric developers.

I don’t plan to replace all my shell scripts with fabfiles, but I’ll seriously consider it next time I need to write a script.