I'm trying to use multiprocessing for compression in python

Question

Here's what I'm trying to do-

import lz4.frame
import multiprocessing as mp

pool = mp.Pool(mp.cpu_count())
chunk_size = 64*1024
#screen is a pygame.Surface object converted to bytes

with lz4.frame.LZ4FrameCompressor(compression_level = 10, auto_flush = True) as compressor:
                compressed = pool.map(compressor.compress, [screen[i : i + chunk_size] for i in range(0, len(screen), chunk_size)])
                compressed = compressor.begin() + b''.join(compressed)
                compressed += compressor.flush()
pool.close()

This works just fine when I use map instead of pool.map... With pool.map, nothing happens. Not even an error...

Also, what is the point of block_size argument in the compress function? I tried different combinations of argument block_size (4 and 5) and chunk_size (64k and 256k), but it didn't seem to make much difference.

Answer 1

Doing multiprocessing inside of a context can be hasardous if it is not fork safe.

In the doc ( https://python-lz4.readthedocs.io/en/stable/intro.html ) I read:

The bindings drop the GIL when calling in to the underlying LZ4 library, and is thread safe.

Have you tried multithreading instead ?

import lz4.frame
from concurrent.futures import ThreadPoolExecutor
import os

nb_chunk = 1024
chunk_size = 64*1024
screen = os.urandom(nb_chunk*chunk_size)

context = lz4.frame.create_compression_context()
chunks = [screen[i : i + chunk_size] for i in range(0, len(screen), chunk_size)]
contexts = [context] * len(chunks)
compressed = lz4.frame.compress_begin(context)
with ThreadPoolExecutor(max_workers=os.cpu_count()) as executor:
    compressed_chunks = executor.map(lz4.frame.compress_chunk, contexts, chunks)
compressed += b''.join(compressed_chunks)
compressed = lz4.frame.compress_flush(context)