One of Python's greatest strengths is the large and diverse set of existing code bases that you can build on to accomplish your goals. Unfortunately, as projects grow, managing your dependencies can become a constant headache.

Unwinding which transitive dependency update now conflicts with other dependencies or is simply broken then somehow preventing that version of the dependency from being used is a common issue that needs to be addressed. Other issues include ensuring the correct system packages are installed so that Python packages can be built and link against system libs properly. Consistently reproducing an installation across more that one machine is surprisingly difficult. Python packages can and do execute arbitrary code on the build machines. Pip does not have a proper dependency resolver. The list goes on and on.

As part of running a continuous integration system that installs Python packages tens of thousands of times per day the OpenStack project has learned how to deal with many of these issues. I will talk about how we use tools from PyPa (like pip constraints) and tools we have built (Bindep, PBR) to reliably perform Python package installs that are consistent and reproducible in the real world.