There is one error in GUI with multiprocess

[szwfive]

The hardware environment is a single 3090. And timing of this error is uncertain. Although the slam process continues later, the stuck GUI cause the memory overflow. This may be the reason why Issue#7 cannot run.

[muskie82]

Hi,

We previously encountered a similar issue randomly in our environment, though it hasn't appeared recently.
If you happen to find a way to reproduce the error deterministically, could you share it with us?

The problem seems to be:

1. PyTorch's multiprocessing communication between the main process and GUI fails for some reason, causing unpredictable GUI crashes.
2. The main process continues to send messages to the queue (q_main2vis), but with the GUI down, these items aren't cleared from GUI process, leading to memory overflow.

This issue occurs only when the GUI is enabled, so it should not be a problem when running in headless mode.

[szwfive]

Thank you for your answer.
I have located the source of the issue, in line 400 of slam frontend. py, "gains=clone_obj (self. issues)". It seems that an unknown error occurred during the copy.deepcopy process. I am trying to find the problem more accurately.

[Tianci-Wen]

I also have the same problem while running this repo in a single GPU 4090. I'm also trying to solve this problem.

[whwh747]

same error

[hnglp]

Thank you for your answer. I have located the source of the issue, in line 400 of slam frontend. py, "gains=clone_obj (self. issues)". It seems that an unknown error occurred during the copy.deepcopy process. I am trying to find the problem more accurately.

Hello, is there any progress on solving the issue? Additionally, I would like to ask if I can run this project with an RTX 3070 (39GB). Regards.

[whwh747]

I'm sorry, I can't resolve it.

[JIAZHAOQIAN]

I modified the function named 'get_latest_queue' in gui_utils.py as follows:

def get_latest_queue(q):
    message = None
    while True:
        try:
            message_latest = q.get_nowait()
            if message is not None:
                del message
            message = message_latest
        except queue.Empty:
            if q.qsize() < 1:
                break
        # zajia: for unsolved bug related to "torch.storage._UntypedStorage"
        except TypeError:
            print("get a torch.storage._UntypedStorage error!")
            break
    return message

I try to capture the error and do not modifiy the 'message' variable. It seems work properly.