> On Dec 18, 2019, at 2:56 AM, Jason H <jh...@gmx.com> wrote: > > I've done something similar in there past. I would generate the images on the > main thread, but then send it to a QThread/QRunnable for saving because I was > using PNG and the save took far longer than anything else. All my cores were > saturated with saves and the QImage generation was 100s of frames ahead. > Fortunately, I had enough RAM to render the sequence. > > Maybe something like that would work better? But I did it all in C++ so no > GIL.
Unfortunately not - my time is spent in the generation/data load, not in the saving. The final images are only 800x1000 pixels, so they save quite quickly. It’s the generation I need to parallelize. Using threads may work - most of the time is spent in either I/O or numpy array operations (which should be C, and so GIL shouldn’t be too much of an issue), but it seems like processes are simply a better option here. That said, I did figure out a workaround - spin any pre-qt web work off to a child process. A bit kludgy, but by doing that it only “breaks” the child process, not the main process, allowing me to spin up as many child processes as I want to run the Qt code without issue. --- Israel Brewster Software Engineer Alaska Volcano Observatory Geophysical Institute - UAF 2156 Koyukuk Drive Fairbanks AK 99775-7320 Work: 907-474-5172 cell: 907-328-9145 > >> Sent: Tuesday, December 17, 2019 at 8:06 PM >> From: "Israel Brewster" <ijbrews...@alaska.edu >> <mailto:ijbrews...@alaska.edu>> >> To: "Jason H" <jh...@gmx.com <mailto:jh...@gmx.com>> >> Cc: pyside@qt-project.org <mailto:pyside@qt-project.org> >> Subject: Re: [PySide] PySide2 + multiprocessing + requests = crash? >> >> >>> On Dec 17, 2019, at 9:56 AM, Jason H <jh...@gmx.com> wrote: >>> >>> QApplication has to be on the main thread. >>> So I would guess (without seeing the crash) that yes, it's bad form. >> >> It *is* on the main thread. Of a child process, true, but the main thread of >> that process. Multiprocessing, not multithreading. Presumably that’s why it >> generally works - as long as I don’t try to make a web request :-) >> >>> >>> Generally when I want to do something like ehat you are doing, I set a >>> singlshot timer to fire (interval: 0) start the event loop (app.exec() or >>> app.exec_() in python) then the event loop wth execute the timer code. >> >> No event loop here. I’m just creating QWidgets, setting contents, and using >> .grab() to save the contents as an image, after which the widgets are >> discarded. The only reason I have a QApplication at all is that you can’t >> create QWidgets without one. >> >> There is no user interaction, no displayed GUI at any point. In fact, I’m >> initializing QApplication with a platform of offscreen. I just want to work >> with multiple QWidgets simultaneously to generate multiple images in >> parallel - thus the need for multiprocessing. >> >>> >>> >>> >>> Sent: Tuesday, December 17, 2019 at 1:13 PM >>> From: "Israel Brewster" <ijbrews...@alaska.edu> >>> To: pyside@qt-project.org >>> Subject: [PySide] PySide2 + multiprocessing + requests = crash? >>> I’ve found what seems to be an odd crash to me. Using PySide2 and >>> multiprocessing works fine to create a QAppliction in a separate process >>> (unless this is considered bad form?) - unless I call requests.get at some >>> point before starting said process, in which case the process crashes (hard >>> crash, not python exception). How might I fix this? >>> >>> A little background: I’m using Qt classes to generate a bunch of images >>> from data files. To speed things along, I launch a process (using >>> multiprocessing.Process) for each image I want to generate, and in each >>> process I create a QApplication, create and fill the widgets I need for the >>> image, and then save the image to a file. This appears to work fine, >>> creating the images in parallel. Today I tried to do a requests call before >>> doing any of the processing (simple get, doesn’t interact with QT in any >>> way), and suddenly my code started crashing. >>> >>> Some simplified sample code that illustrates the issue: >>> >>> import time >>> import multiprocessing >>> import sys >>> from PySide2.QtWidgets import QApplication >>> import requests >>> >>> >>> def test_wait(): >>> print("Starting proc") >>> try: >>> QApplication(sys.argv + ['-platform', 'offscreen']) >>> except Exception as err: >>> print("oops", err) >>> >>> print("Sleeping now...") >>> time.sleep(10) >>> >>> if __name__ == "__main__": >>> proc = multiprocessing.Process(target=test_wait) >>> proc.start() >>> proc.join() >>> print("Awake again. Let's try another:") >>> >>> res = requests.get('https://www.google.com <https://www.google.com/> >>> <https://www.google.com/ <https://www.google.com/>>') # or whatever your >>> favorite is >>> res.text # Get the response, close it out >>> >>> proc = multiprocessing.Process(target=test_wait) >>> start = time.time() >>> proc.start() >>> proc.join() >>> >>> # The process should sleep for 10 seconds, so if we got here in less >>> than 9, we died. >>> if time.time() - start < 9: >>> print("Never slept :(") >>> else: >>> print("We lived!") >>> >>> --- >>> Israel Brewster >>> Software Engineer >>> Alaska Volcano Observatory >>> Geophysical Institute - UAF >>> 2156 Koyukuk Drive >>> Fairbanks AK 99775-7320 >>> Work: 907-474-5172 >>> cell: 907-328-9145 >>> _______________________________________________ PySide mailing list >>> PySide@qt-project.org <mailto:PySide@qt-project.org> >>> https://lists.qt-project.org/listinfo/pyside >>> <https://lists.qt-project.org/listinfo/pyside> >>> <https://lists.qt-project.org/listinfo/pyside >>> <https://lists.qt-project.org/listinfo/pyside>>
_______________________________________________ PySide mailing list PySide@qt-project.org https://lists.qt-project.org/listinfo/pyside
