This talk explores how Python manages object lifetimes through references, and why objects sometimes remain in memory longer than expected. While Python’s memory model is often described as simple, its real behavior — especially in the presence of caching and reference cycles — can lead to surprising and hard-to-debug issues.

The talk is structured as a narrative in three parts: strong references, weak references, and the garbage collector. First, we take a look at Python’s object model and reference counting, using simple, visual examples to demonstrate how references work. Next, we introduce weak references and show how they differ from regular references, including practical use cases and common misconceptions. Finally, we examine the garbage collector and explain why it is needed in Python on top of reference counting.

The concepts are grounded in a real-world debugging story involving functools.lru_cache, where a small optimization led to unexpected memory growth.

The goal of this talk is to provide a clear and practical mental model of Python’s memory behavior, helping developers understand not just how things work, but why seemingly correct code can lead to subtle memory issues.