Recently, Meta introduced the concept of immortal objects to Python. The basic idea is to have objects hold a flag stating, "Please don't delete me!"
When a garbage collector does the rounds, it ignores all objects having this flag. And this small change has a big impact on its current and future capabilities.
Objects in Python are augmented with a reference counter. This counter tells you the number of incoming references to an object. A garbage collector iteratively deletes objects which have a reference count equal to zero.
1. Garbage Collection with Reference Counting
Contrast this with Java or Golang, which runs a mark and sweep (Depth First Search) algorithm to find dead objects.
2. Garbage Collection with Mark and Sweep
The problem with reference counts is sharing objects across threads. When an object is referenced, the counter increments. This operation is not thread-safe, and needs either of two approaches to work:
- Copy on write (create an object copy when adding a reference to it)
- Global interpreter lock (Take a lock before making changes)
Both of these approaches have downsides. The first is wasteful, the second is slow.
Instead, Meta's changes help define objects as Immutable. This means that the reference count variables of these objects are also immutable.
Now, the garbage collector can ignore these objects.
3. Garbage Collection with Immortal Objects
This simple change allows objects to be shared efficiently across threads. The memory savings through this change at Meta are over 30%!
If you want to know more about system design and software engineering, try our system design course at InterviewReady.
Cheers!
Reference: https://engineering.fb.com/2023/08/15/developer-tools/immortal-objects-for-python-instagram-meta