2 years ago

Category: Kubernetes, Docker, artificial-intelligence

It was a dark and stormy day indeed when, by a stroke of luck, our intrepid heroes stumbled across an entry in the pip changelog: Maintainers are preparing to release pip 20.3, with the new resolver on by default. Confusion and curiosity struck their hearts!

We use tools like Renovate to ensure we don’t need to micromanage updates to our dependencies, pre-commit to run an ever-growing litany of linters, and CircleCI to automatically run those linters, tests, and everything involved with the continuous delivery of our code; we build applications and libraries (and scripts which act somewhere in between!), run our code on Knative and Kubernetes and some serverless platforms, write code in Python which can interface with our other code in C++, Elixir, Java, …

Since the AI teams have a number of codebases spread across several teams with varying levels of complexity and importance to our users, it was important to carefully plan how to make our changes.

It was most important to get these issues solved first for our most important systems in case our timeline estimate was wildly off and the pip-ageddon happened before we were ready.