search

python-pillow / Pillow

Python Imaging Library (Fork)
https://python-pillow.org
Other
12.29k stars 2.23k forks source link

Memory leak when opening images #7961

Open aIligat0r opened 7 months ago

aIligat0r commented 7 months ago

Hello! I have not found a solution to this problem. If there was a decision, then I apologize in advance.

When opening images, the amount of memory is constantly increasing. The library somehow saves the open images. I'm opening an image in context.

Simulated leak:

import os
import tracemalloc

import psutil
from PIL import Image

def main():
    with open("/tmp/github.png", "rb") as img_file:
        with Image.open(fp=img_file) as img:
            pass

if __name__ == '__main__':
    print("PIL version:", PIL.__version__)
    tracemalloc.start()
    pid = os.getpid()
    process = psutil.Process(pid)
    for i in range(100):
        main()
        snapshot = tracemalloc.take_snapshot()
        stats = snapshot.statistics("lineno")
        for s in stats[:10]:
            print(s.size, s.traceback)
        memory_info = process.memory_info()
        print(f'RAM usage: {memory_info.rss / 1000 / 1000} MB')
        print(f"Replay: {i+1}\n" + "*" * 70)
    tracemalloc.stop()

Out:

PIL version: 10.3.0

882738 <frozen importlib._bootstrap_external>:729
34816 /usr/lib/python3.11/tracemalloc.py:67
30464 /usr/lib/python3.11/tracemalloc.py:505
26400 /usr/lib/python3.11/tracemalloc.py:558
26160 /usr/lib/python3.11/tracemalloc.py:498
7727 <frozen importlib._bootstrap>:241
5286 <frozen importlib._bootstrap_external>:128
4788 /usr/lib/python3.11/enum.py:532
4408 /usr/lib/python3.11/tracemalloc.py:534
4228 /home/user/venv/lib/python3.11/site-packages/PIL/PngImagePlugin.py:700
RAM usage: 21.614592000000002 MB
Replay: 4
==================================================================
882738 <frozen importlib._bootstrap_external>:729
34816 /usr/lib/python3.11/tracemalloc.py:67
30296 /usr/lib/python3.11/tracemalloc.py:505
26016 /usr/lib/python3.11/tracemalloc.py:498
25536 /usr/lib/python3.11/tracemalloc.py:193
7727 <frozen importlib._bootstrap>:241
5286 <frozen importlib._bootstrap_external>:128
4788 /usr/lib/python3.11/enum.py:532
4392 /usr/lib/python3.11/tracemalloc.py:534
4228 /home/user/venv/lib/python3.11/site-packages/PIL/PngImagePlugin.py:700
RAM usage: 21.614592000000002 MB
Replay: 5
==================================================================
882738 <frozen importlib._bootstrap_external>:729
34816 /usr/lib/python3.11/tracemalloc.py:67
30352 /usr/lib/python3.11/tracemalloc.py:505
26016 /usr/lib/python3.11/tracemalloc.py:498
25536 /usr/lib/python3.11/tracemalloc.py:193
7727 <frozen importlib._bootstrap>:241
5286 <frozen importlib._bootstrap_external>:128
4788 /usr/lib/python3.11/enum.py:532
4392 /usr/lib/python3.11/tracemalloc.py:534
4228 /home/user/venv/lib/python3.11/site-packages/PIL/PngImagePlugin.py:700
RAM usage: 21.745664 MB
Replay: 6
==================================================================

...

880946 <frozen importlib._bootstrap_external>:729
34880 /usr/lib/python3.11/tracemalloc.py:67
30464 /usr/lib/python3.11/tracemalloc.py:505
26112 /usr/lib/python3.11/tracemalloc.py:498
15812 /home/user/venv/lib/python3.11/site-packages/PIL/PngImagePlugin.py:192
13248 /usr/lib/python3.11/tracemalloc.py:193
9344 /usr/lib/python3.11/tracemalloc.py:558
7727 <frozen importlib._bootstrap>:241
5286 <frozen importlib._bootstrap_external>:128
4788 /usr/lib/python3.11/enum.py:532
RAM usage: 21.876736 MB
Replay: 98
==================================================================
880946 <frozen importlib._bootstrap_external>:729
34880 /usr/lib/python3.11/tracemalloc.py:67
30408 /usr/lib/python3.11/tracemalloc.py:505
26064 /usr/lib/python3.11/tracemalloc.py:498
19488 /usr/lib/python3.11/tracemalloc.py:193
15871 /home/user/venv/lib/python3.11/site-packages/PIL/PngImagePlugin.py:192
7727 <frozen importlib._bootstrap>:241
5286 <frozen importlib._bootstrap_external>:128
4788 /usr/lib/python3.11/enum.py:532
4408 /usr/lib/python3.11/tracemalloc.py:534
RAM usage: 21.876736 MB
Replay: 99
==================================================================
880946 <frozen importlib._bootstrap_external>:729
34880 /usr/lib/python3.11/tracemalloc.py:67
30464 /usr/lib/python3.11/tracemalloc.py:505
26112 /usr/lib/python3.11/tracemalloc.py:498
15930 /home/user/venv/lib/python3.11/site-packages/PIL/PngImagePlugin.py:192
12512 /usr/lib/python3.11/tracemalloc.py:558
10032 /usr/lib/python3.11/tracemalloc.py:193
7727 <frozen importlib._bootstrap>:241
5286 <frozen importlib._bootstrap_external>:128
4788 /usr/lib/python3.11/enum.py:532
RAM usage: 21.876736 MB
Replay: 100
==================================================================

It looks like a memory leak in the plugin PngImagePlugin.py

If there is a solution how to get around this problem in tasks where you have to open a lot of images?

I have previously encountered an error in Webp formats. The leak was in the WebPImagePlugin plugin. I solved this by changing the value of the variable HAVE_WEBPANIM (PIL._webp.HAVE_WEBPANIM) to False. But now I'm facing a problem in PngImagePlugin

Yay295 commented 7 months ago

It looks like it thinks the errors might be coming from lines 192 and 700?

https://github.com/python-pillow/Pillow/blob/955c5dac03bd4b59281d00ffad345aa28e79310f/src/PIL/PngImagePlugin.py#L188-L192

https://github.com/python-pillow/Pillow/blob/955c5dac03bd4b59281d00ffad345aa28e79310f/src/PIL/PngImagePlugin.py#L700-L702

radarhere commented 7 months ago

If I increase the number of loops to 1000, I find that the memory drops down again at a certain point. Since it is not continuously increasing, I don't think it is a leak.

You might like to read #7935, in particular https://github.com/python-pillow/Pillow/issues/7935#issuecomment-2031804237

Pillow's memory allocator doesn't necessarily release the memory in the pool back as soon as an image is destroyed, as it uses that memory pool for future allocations. See Storage.c (https://github.com/python-pillow/Pillow/blob/main/src/libImaging/Storage.c#L310) for the implementation.

Yay295 commented 7 months ago

If I increase the number of loops to 1000, I find that the memory drops down again at a certain point.

It would probably be good to add gc.collect() (import gc) to the end of each loop. It might just not be running that often.

radarhere commented 5 months ago

It would probably be good to add gc.collect() (import gc) to the end of each loop. It might just not be running that often.

I added gc.collect() in, but it doesn't make an obvious difference.

It looks like it thinks the errors might be coming from lines 192 and 700?

I see line 1083 on my machine. This would be easier to discuss if the original image could be uploaded here.

I have previously encountered an error in Webp formats. The leak was in the WebPImagePlugin plugin. I solved this by changing the value of the variable HAVE_WEBPANIM (PIL._webp.HAVE_WEBPANIM) to False.

If I test a WebP image with your above code, I again find that the memory drops down again at a certain point.

jeethesh-pai commented 3 weeks ago

If I increase the number of loops to 1000, I find that the memory drops down again at a certain point.

It would probably be good to add gc.collect() (import gc) to the end of each loop. It might just not be running that often.

This saved me a lot. I encountered a similar problem where I load HD images into memory (35MiB per Image), saw that the RAM quickly rose up to 1 GiB within short span of time. I solved this by assigning img=None and gc.collect() at the end of the every iteration in the loop. For some reason garbage collector does not seem to collect garbage variables quickly enough in multithreaded/multiprocessing setting