Leer The GIL and Its Memory Implications

Veeg om het menu te tonen

The Global Interpreter Lock (GIL) is a mutex that allows only one thread to execute Python bytecode at a time. It exists to protect CPython's reference counting mechanism from race conditions. Understanding the GIL explains why threading does not help with CPU-bound work and why it affects memory access patterns.

Why the GIL Exists

Reference counting increments and decrements are not atomic operations. Without the GIL, two threads modifying the same object's reference count simultaneously could corrupt it, leading to premature deallocation or memory leaks:


              1234567891011121314151617181920
            
import sys
import threading

# Demonstrating that reference counts are consistent under the GIL
shared_data = {"revenue": 142500.0, "costs": 98000.0}

def read_data(iterations):
    for _ in range(iterations):
        # Safe – GIL ensures reference count operations are atomic
        value = shared_data["revenue"]
        _ = value + 1

threads = [threading.Thread(target=read_data, args=(100000,)) for _ in range(4)]
for thread in threads:
    thread.start()
for thread in threads:
    thread.join()

print(f"Reference count after threading: {sys.getrefcount(shared_data)}")
print("No corruption – GIL protected the reference count")

The GIL and Memory-Bound Work

For I/O-bound tasks, the GIL is released during blocking calls – threads can run concurrently while waiting for network or disk. For CPU-bound tasks, the GIL prevents real parallelism:


              1234567891011121314151617181920212223242526
            
import threading
import time

# CPU-bound work – GIL prevents parallel execution
def compute_sum(limit):
    total = sum(value ** 2 for value in range(limit))
    return total

# Sequential
start_time = time.time()
compute_sum(300000)
compute_sum(300000)
sequential_time = time.time() - start_time

# Threaded – NOT faster due to GIL
start_time = time.time()
thread_a = threading.Thread(target=compute_sum, args=(300000,))
thread_b = threading.Thread(target=compute_sum, args=(300000,))
thread_a.start()
thread_b.start()
thread_a.join()
thread_b.join()
threaded_time = time.time() - start_time

print(f"Sequential: {sequential_time:.2f}s")
print(f"Threaded:   {threaded_time:.2f}s")  # Similar or slower due to GIL overhead

Memory and the GIL: Thread-Safe Reference Counting

The GIL makes simple attribute reads and writes on Python objects thread-safe at the bytecode level. However, compound operations — check-then-act sequences — are not atomic:


              1234567891011121314151617
            
import threading

# Unsafe compound operation – not protected by GIL alone
shared_counter = {"value": 0}

def increment(iterations):
    for _ in range(iterations):
        # Not atomic – read + increment + write are separate bytecodes
        shared_counter["value"] += 1

threads = [threading.Thread(target=increment, args=(10000,)) for _ in range(4)]
for thread in threads:
    thread.start()
for thread in threads:
    thread.join()

print(f"Expected: 40000, Got: {shared_counter['value']}")  # May not be 40000

Fix compound operations with a threading.Lock():


              1234567891011121314151617
            
import threading

shared_counter = {"value": 0}
lock = threading.Lock()

def safe_increment(iterations):
    for _ in range(iterations):
        with lock:
            shared_counter["value"] += 1

threads = [threading.Thread(target=safe_increment, args=(10000,)) for _ in range(4)]
for thread in threads:
    thread.start()
for thread in threads:
    thread.join()

print(f"Expected: 40000, Got: {shared_counter['value']}")  # Always 40000

GIL and Memory: Key Points

The GIL makes reference counting thread-safe, preventing memory corruption;
It does not make compound Python operations atomic;
For CPU-bound parallelism, use multiprocessing – each process has its own GIL and its own memory space;
For I/O-bound concurrency, use threading or asyncio – the GIL is released during blocking I/O.

Was alles duidelijk?

Bedankt voor je feedback!

Sectie 4. Hoofdstuk 1

Vraag AI

Vraag wat u wilt of probeer een van de voorgestelde vragen om onze chat te starten.

Sectie 4. Hoofdstuk 1