GIL

A summary about the GIL (Global Interpreter Lock) in Python. Based on information from this very nice article.

What is it ?

A lock inside the main evaluation loop which makes Python code thread-safe by nature
Allow only one OS thread to execute Python bytecode at any given time
Protects mutable data sructures in user code but also from Python: e.g. reference counting

Python Threads ?

Python has support for threads, but due to the GIL, it won’t speed up your application
I/O operations do not hold the GIL 1. Release the GIL 2. Do the I/O operation (write, recv, accept, …) 3. Acquire the GIL

What is not running in the GIL ?

I/O Operations
C code or extension

Evaluation loop

This is a high level view of the Python evaluation loop:

while true
  check if other threads requested the GIL (eval_breaker)
    if yes, release the GIL and waits again
  get the next bytecode instruction
  run the instruction

Switch interval

Before entering the instruction loop, the thread will try to acquire the GIL
If it’s available, the thread continues and runs
If it is not, it waits until it is released
By default, it waits for 5ms (switch interval)
If the GIL is not released during that time, it sets the eval_breaker
The thread holding the GIL will suspend execution
The waiting thread can now acquire the GIL and run

This means for I/O threads, if there is another CPU-bound thread running, it might take 5ms to run again after each I/O call!