pull request about issue #2062 (car_racing.py memory leaking issue )

[jackyoung96]

Pull request about issue #2062

There is memory leaking issue caused by pyglet module and solved with simply delete after drawing vertexs

I confirmed with e-mail from @praveenVnktsh

[SwapnilPande]

Hi @Jaekyung-Cho,

I've been debugging the same issue. Looks like we need to add vl.delete() after vl.draw in function render_indicators (line 592). It looks like the memory consumption is still growing after this change, but much much slower.

[jackyoung96]

Hi @Jaekyung-Cho,

I've been debugging the same issue. Looks like we need to add vl.delete() after vl.draw in function render_indicators (line 592). It looks like the memory consumption is still growing after this change, but much much slower.

Exactly. Memory leaking still remain, but I cannot find where it occur. Though slower leaking issues doesn't matter in my case.

[SwapnilPande]

Looks like this is an issue with Box2D, I was able to isolate the memory leak down to creating the tiles in _create_track. Despite calling DestroyBody on all of the tiles in _destroy, Box2D does not properly destroy the tiles sometimes. This memory leak can be reproduced with this minimum code:

import Box2D
from Box2D.b2 import fixtureDef, polygonShape

import psutil
import os

def memory_used():
    return psutil.Process(os.getpid()).memory_info().rss * 1e-6  # To megabyte


world = Box2D.b2World((0,0))


road1_l = (0, 0)
road1_r = (0, 1)
road2_l = (1, 0)
road2_r = (1, 1)
vertices = [road1_l, road1_r, road2_r, road2_l]

fd_tile = fixtureDef(
                shape = polygonShape(vertices=
                    [(0, 0),(1, 0),(1, -1),(0, -1)]))

fd_tile.shape.vertices = vertices

for i in range(1000000):
    t = world.CreateStaticBody(fixtures=fd_tile)
    world.DestroyBody(t)

    if i % 10 == 0:
        print('Ram Used: %f' % memory_used())

Looks like it's time to submit an issue to pybox2d. :D

The slow leaking doesn't matter in my case either, though it'd be nice to have this issue resolved. Would you be able to add the second vl.delete() statement to make the PR complete for the gym codebase?

[jackyoung96]

Looks like this is an issue with Box2D, I was able to isolate the memory leak down to creating the tiles in _create_track. Despite calling DestroyBody on all of the tiles in _destroy, Box2D does not properly destroy the tiles sometimes. This memory leak can be reproduced with this minimum code:

import Box2D
from Box2D.b2 import fixtureDef, polygonShape

import psutil
import os

def memory_used():
    return psutil.Process(os.getpid()).memory_info().rss * 1e-6  # To megabyte


world = Box2D.b2World((0,0))


road1_l = (0, 0)
road1_r = (0, 1)
road2_l = (1, 0)
road2_r = (1, 1)
vertices = [road1_l, road1_r, road2_r, road2_l]

fd_tile = fixtureDef(
                shape = polygonShape(vertices=
                    [(0, 0),(1, 0),(1, -1),(0, -1)]))

fd_tile.shape.vertices = vertices

for i in range(1000000):
    t = world.CreateStaticBody(fixtures=fd_tile)
    world.DestroyBody(t)

    if i % 10 == 0:
        print('Ram Used: %f' % memory_used())

Looks like it's time to submit an issue to pybox2d. :D

The slow leaking doesn't matter in my case either, though it'd be nice to have this issue resolved. Would you be able to add the second vl.delete() statement to make the PR complete for the gym codebase?

Oh! I think I misunderstand your mention. I'm so sorry about that. I add vl.delete() in render_indicators function and make PR now. Thank you for your help.

[hirekk]

vl.delete() should also be added at the end of the render_road method (line 530), since the memory leak still occurs. As far as I can tell, this fixes the issue completely.

[yoavalon]

This solution works well for me. Memory remains low,

[SlipknotTN]

Me too

[xeviknal]

Any updates in this one? Do you think this PR is any close to be merged?

[xeviknal]

I've been able to reproduce the issue. Once the PR is applied, the memory leak is gone or at least not exploding.

[jackyoung96]

I've been able to reproduce the issue. Once the PR is applied, the memory leak is gone or at least not exploding.

Thanks for reproducing instead of me!!

[xeviknal]

@pzhokhov @zuoxingdong I see you've been contributing quite a lot last months. Do you know what we can do to move this forward? are you maintainers? I am guessing this issue is making people to start forking. I'd be pretty good for this community to approach this nasty bug. thank you

A release with the fixing would be appreciated too 👼