Understanding Python Memory Management

Chelsea Dole

Data Engineer

​

What is “memory”, and what manages it in Python?

• RAM (“Random Access Memory”) is used by applications on your computer for short-term, fast-access data storage

​

• In CPython, the “Python memory manager” handles the allocation and deallocation of memory used for objects in your programs
• Creating a variable
• Utilizing a data structure, like a dictionary or list
• “Call stack frames”

“test”

TYPE: int, VALUE: 1000, REF COUNT: 1

• Variable — a “label” referencing an object in memory

“test”

TYPE: int, VALUE: 1000, REF COUNT: 1

“another_test”

TYPE: int, VALUE: 1000, REF COUNT: 2

• Variable — a “label” referencing an object in memory

• ​
• ​
• ​

• ​
• ​

So, what was that “ref count” thing?

• Reference count — Python’s way of tracking the number of “labels” referencing an object in memory.

​

Increasing Ref Count:

Decreasing Ref Count:

1000

Garbage Collection!

🎉

🎉

🚛

🚛

• Garbage Collection — an automated memory management technique used in the source code of many languages (including Python!) to free up memory that is no longer in use

• Static memory allocation
• Memory manually managed by the developer, and allocated at compile time
• NO garbage collection

​

​

Source code is faster and uses less memory. Bug prone for developers.

Source code is more complex, and uses more memory. Easier for developers.

​

• Dynamic memory allocation
• Memory never managed directly by the developer, and allocated by the Python Memory Manager at runtime
• Garbage collection!

However, ref counting isn’t reliable enough to be Python’s exclusive method of garbage collection

1. Python keeps track of all objects with ref count > 0 by age in three “generation buckets”
2. Once a bucket reaches a certain size, a “Mark and Sweep” algorithm is run on it, and every generation below it
1. ✍️ Check all objects in the generation bucket, and “mark” all reachable objects
2. 🧹Delete everything that’s not “marked”
3. 🎓Promote all surviving objects to the next generation

​

​

​

Solution: generational “Mark and Sweep” algorithm

1. Reference counting

​

• Generational “Mark and Sweep”algorithm

How a language chooses to manage memory is a major functional and philosophical choice

​

​

• Speed, or ease of use for developers?
• Clarity of source code, or clarity of language syntax?

Python chooses to maintain a complex and relatively slow memory management system internally, because it enables Python to be simple, beautiful, and readable for developers.

Thank you!

Chelsea Dole

www.chelseadole.com