Python's import system runs eagerly: every import statement executes the module you are importing from top to bottom, and this includes executing all the nested imports. For command-line tools like Django's management commands, this can result in a multi-second wait. For our project, this resulted in 10+ second wait times on every command, including every server restart, due to a few bad imports scattered around the codebase.

In the talk, we'll briefly explain how the import system works, how to find imports that take a long time, and explain why the slowest import was so slow.

We'll then look at how using lazy imports reduced our management commands from 10 seconds to less than 3 seconds, but also how our codebase ended up looking worse as a result. We'll also look at how to ensure we don't accidentally make end users pay for lazy imports during the request/response cycle, and how we made sure it's not possible to regress by adding a lint check.

Finally, we'll look at how, in Python 3.15, all of this will become much easier thanks to PEP 810: Explicit lazy imports.