EmulationStation freezes after game-select event triggered and video snap plays

obrie572

@mitu Thanks for helping take a look

I installed emulationstation-dev and am able to reproduce the same.

$ sudo $HOME/RetroPie-Setup/retropie_packages.sh emulationstation-dev _binary_

$ emulationstation --help
EmulationStation, a graphical front-end for ROM browsing.
Written by Alec "Aloshi" Lofquist.
Version 2.11.0rp-dev, built Feb 13 2022 - 16:52:03

mitu

I've been testing with the dev version, but I still can't reproduce the freeze/crash. Maybe it's video related ? Can you try to remove the game-select event script and see if it still happens ?

obrie572

I removed the game-select event and the freeze no longer occurs. Here's the output of a ffprobe for an example video:

Input #0, mov,mp4,m4a,3gp,3g2,mj2, from '/home/pi/.emulationstation/downloaded_media/arcade/videos/88games.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41
    title           : 88games
    artist          : EmuMovies
    album_artist    : Circo
    album           : MAME Video Snaps
    date            : 2015
    encoder         : Lavf57.56.101
    comment         : http://emumovies.com
    genre           : Video Snaps
  Duration: 00:00:30.05, start: 0.000000, bitrate: 477 kb/s
    Stream #0:0(und): Video: h264 (High) (avc1 / 0x31637661), yuv420p, 640x480 [SAR 1:1 DAR 4:3], 406 kb/s, 30 fps, 30 tbr, 15360 tbn, 60 tbc (default)
    Metadata:
      handler_name    : VideoHandler
    Stream #0:1(und): Audio: mp3 (mp4a / 0x6134706D), 24000 Hz, stereo, fltp, 64 kb/s (default)
    Metadata:
      handler_name    : SoundHandler

I also tried increasing the video delay in the theme from 3 seconds to 5 seconds, but still encountered the issue.

I can see if I can add some additional debugging into a custom build of ES to see if I can gather more information. Understanding the cause of the script sometimes failing with a non-zero exit code might also lead to something insightful.

obrie572

After a little debugging, it looks like this is a race condition between the system() call and the SIGCHLD signal handler.

The SIGCHLD signal handler is preventing system() calls from executing properly. I believe this thread accurately describes the issue I'm seeing: https://stackoverflow.com/questions/17550217/linux-system-sigchld-handling-multithreading

I commented out the line in ES that adds the SIGCHLD signal handler for the omx player and I no longer experience the freeze. Of course, I have zombie child processes that need to get reaped -- but this at least confirmed the theory.

mitu

Nice find !
It does seem to be race, perhaps the SIGCHLDs are queued and the system's signal doesn't get delivered until omxplayer is finished playing (forcibly) - that would explain why you get all the queued debug logging when omxplayer is stopped.
Looks like using system for the scripting events is not safe, perhaps a similar approach via fork needs to be implemented.

obrie572

👍 I experimented with a few potential solutions. I'll take another look at the fork approach for the scripting events. I originally tried that early on and didn't have success though I may not have been doing it correctly. I'll put together a branch for that later today.

I believe I have an alternative working solution here: https://github.com/RetroPie/EmulationStation/compare/master...obrie:game-select-omx-freeze

That solution changes the signal handler to have the video component only reap its own pids. Not sure how well that follows best practices, though.

obrie572

I've come up with a better, simpler solution: https://github.com/RetroPie/EmulationStation/compare/master...obrie:game-select-omx-freeze-alt

The problem is that the SIGCHLD handler is not taking into account that the handler may be called only once when there are multiple child processes to be reaped. In this scenario, the following happens:

The system gamelist view is rendered and the first game is selected
After 3 seconds, the emulationstation process is forked, the SIGCHLD handler is registered, and the video starts playing
When you attempt to navigate down to the next game, both a game-select event is triggered and the video is stopped
Under the right conditions, these 2 events can happen effectively simultaneously resulting in 2 child processes that need to be reaped -- the forked omx process and the system() call to the select-game script
When this happens, a single call is made to the SIGCHLD handler

The VideoPlayerComponent's SIGCHLD handler only ever expects there to be a single child reaped. In this case, however, there are multiple child processes to be reaped.

So the solution here is fairly straightforward: introduce a somewhat standard solution of reaping child processes until there are no more to be reaped.

After this change is made, the freezing issue goes away 🎉

@mitu Let me know if you think this is ready for a PR and I'll go ahead and submit one.

mitu

It looks ok and it's simpler than replacing system in the runSystemCommand function.

obrie572

PR created: https://github.com/RetroPie/EmulationStation/pull/790

mitu

@obrie572 The modifications have been merged by @pjft, so they they should be available in the -dev branch right now, until promotod to the stable branch.

Thanks again for the bug report and the fix!