crash

[botiapa]

mpvpaper works for about 30 seconds before crashing with the following error:

[ffmpeg] AVHWDeviceContext: Failed to query surface attributes: 20 (the requested function is not implemented).
[libmpv_render/vaapi-egl] failed to retrieve libavutil frame constraints
Cannot load libcuda.so.1

Playing: /usr/share/sddm/themes/aerial/playlists/./apple_movies/b7-2.mov
 (+) Video --vid=1 (*) (h264 1920x1080 30.000fps)
VO: [libmpv] 1920x1080 yuv420p
wl_display@1: error 1: invalid arguments for zwp_linux_buffer_params_v1@1616.add

Let me know if you need more information regarding the error.

[GhostNaN]

Sorry, I don't have access to my PC for another week or so. But the first thing I would try is a different version of wlroots. So if you have 0.14.1, try the git version.

If that doesn't solve it then I think there might be something wrong with the Wayland protocols I provided. Can't say for sure for now. I'll let you know if I need more info later.

[tmpm697]

I can confirm this issue as it happens every time I used mpvpaper in background for a while. Steps:

1. Start pc, I use mpvpaper to run a .mp4 file in background
2. After I start the pc, the video will be put in pause state -- echo '{ "command": ["get_property", "pause"] }' | socat - /tmp/mpv-socket
3. Run the pc for a while
4. Observer lag input at some points and constantly high ram usage
5. Check ps/top for high memory usage does not show mpvpaper (sometime does)
6. Multiple force kill in dmesg as pc crashed, mpvpaper is the last entry in dmesg
   [25542.067107] Out of memory: Killed process 1417 (mpvpaper) total-vm:1919808kB, anon-rss:19220kB, file-rss:0kB, shmem-rss:9108kB, UID:1000 pgtables:984kB oom_score_adj:0

System:

os: archlinux up-to-date
wm: sway version 1.7-c1725c8 (Apr  9 2022, branch 'community/packages/sway') (from arch official repo)
wlroots: 0.15.1-3 (from arch official repo)

[GhostNaN]

Your issue is a separate issue. I could not find any memory leak that you mentioned with valgrind. Besides "total-vm:1919808kB" (1.9gB) and "anon-rss:19220kB"(19mB) is not out of the ordinary memory usage mpv could use.

I'm not saying mpvpaper is not a fault, but I have some questions:

1. "After I start the pc, the video will be put in pause state" Are you running mpvpaper just to pause it? Why not just use swaybg with an image?
2. "Run for a while", like approximately how long is that? I run my pc at most for 16 hours a day. So perhaps I would never see this issue.
3. "Check ps/top for high memory usage does not show mpvpaper". How much memory is mpvpaper using? As an example, I use htop to check memory usage. And sometimes I've seen memory usage as high as 5G in VIRT and 1.1G in RES.
4. How much memory does your system have?
5. Are you sure there isn't anything else leaking memory? OOM could be a result of another program leaking and Linux is killing mpvpaper in order to reclaim memory back.

Thanks for the report.

[tmpm697]

I don't at pc right now but here're some more info:

1. I pause at startup of pc but occasionally I'll run it. As mostly I'll run it when in less busy workflow or more revealing screen's spaces.
2. My session often around 10-12hrs but it rams up very quick >90% and I have 16G of ram.
3. Does not at pc rn but mpvpaper seems to always top 3 high ram usage.
4. 16GB
5. I use only firefox-beta from aur, tmux, alacritty.

Sway starts to kill programs gradually and the last in dmesg is mpvpaper. I don't assume the issue is from mpvpaper but it's likely as it's last killed.

Mp4 file is an 4k video and around 4GB.

How can I debug this issue?

[GhostNaN]

16GB of RAM, I guess that makes 2 of us and should be more than plenty. "always top 3 high ram usage", As I said before mpv / mpvpaper, especially with a 4k video, will use quite a bit of RAM so it's not too surprising it's in your top 3. I'm also sure you are using software(CPU) instead of hardware(GPU) rendering. As that will also use more RAM.

As for debugging, if you have the time I'm curious if just normal mpv has this issue. It might be a upstream issue. If you don't, at least tell me approximately how much RAM does mpvpaper use at the start and after "a while".

The only time I see a major increase in RAM personally is when initially playing a video(around a minute as the buffers fill) or when the next video plays. Although it can also go back down with a new video playing afterward, depending on the video.

Lastly what are your terminal options? Are you utilizing either a "pauselist" or "stoplist" text file located in ~/.config/mpvpaper/

[tmpm697]

Ram is at 93% and below is checked info: This is from free -h:

free -h
               total        used        free      shared  buff/cache   available
Mem:            15Gi       3.1Gi       585Mi        10Gi        11Gi       1.1Gi
Swap:           31Gi       4.6Gi        27Gi

htop shows mpvpaer has VIRT at 1854M and RES at 67892 (67M)

Note that I start mpvpaper through sway exec like: swaymsg exec 'mpvpaper -v -o <..>'

I installed latest mpv from arch official repo:

mpv 0.34.1-dirty Copyright © 2000-2021 mpv/MPlayer/mplayer2 projects
 built on UNKNOWN
FFmpeg library versions:
   libavutil       57.17.100
   libavcodec      59.18.100
   libavformat     59.16.100
   libswscale      6.4.100
   libavfilter     8.24.100
   libswresample   4.3.100
FFmpeg version: n5.0

In mpv.conf, I've set these settings in hope that mpv will use gpu when plays video and awares wayland context:

hwdec=vaapi
vo=gpu
gpu-context=wayland

As I stated, I pause mpvpaer at start as below (when mpv-socket is avail):

echo 'cycle pause' | socat - /tmp/mpv-socket

I often have ram at >90 with firefox-beta, tmux on alacritty, and some alacritty terminal windows. But as you can see from free command above, most of them are cached data -- probably firefox just use them for speed up pages.

I can't reliable produce this issue but in high workload it happens sometimes.

[GhostNaN]

Wow, your mpvpaper uses less RAM than me! I guess I was wrong that you were using software rendering. If you want you can "free up" your memory from cache with this: sync && sudo sh -c "echo 3 > /proc/sys/vm/drop_caches" Although it's kind of unnecessary as Linux will free up cache as it needs.

"in high workload it happens sometimes", then maybe it's the program/s producing the high RAM usage? For example, if I compile let's say firefox and use /tmp as my build directory my RAM will quickly be eaten up to the point of crashing the system.

If you encounter this problem again and see mpvpaper is using a lot of RAM, something like: pmap -X 871 > mpvpaper_pmap.txt where 871 is pidof mpvpaper. Then upload the text file, that would at least narrow down a few possibilities easily.

[tmpm697]

Thanks, I'll try to get that log when I see some symptoms is about to crash.

[tmpm697]

I've just got a crash, this time swaywm crashed also:

[ 3565.798390] Out of memory: Killed process 95199 (dbus-daemon) total-vm:8544kB, anon-rss:444kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:56kB oom_score_adj:200
[ 3565.838660] Out of memory: Killed process 779 ((sd-pam)) total-vm:104780kB, anon-rss:52kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:92kB oom_score_adj:100
[ 3568.695566] Out of memory: Killed process 778 (systemd) total-vm:18080kB, anon-rss:1288kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:72kB oom_score_adj:100
[ 3601.516532] Out of memory: Killed process 95445 (rclone) total-vm:60816kB, anon-rss:4280kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:88kB oom_score_adj:200
[ 3602.473405] Out of memory: Killed process 95441 (hstdb) total-vm:13604kB, anon-rss:5464kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:64kB oom_score_adj:200
[ 3607.472980] Out of memory: Killed process 95444 (rclone) total-vm:761828kB, anon-rss:5980kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:144kB oom_score_adj:200
[ 3614.480054] Out of memory: Killed process 95471 ((lipmenud)) total-vm:17792kB, anon-rss:1900kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:76kB oom_score_adj:200
[ 3620.882254] Out of memory: Killed process 95422 ((sd-pam)) total-vm:171628kB, anon-rss:2564kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:104kB oom_score_adj:100
[ 3635.799255] Out of memory: Killed process 95421 (systemd) total-vm:17792kB, anon-rss:1204kB, file-rss:0kB, shmem-rss:0kB, UID:1000 pgtables:76kB oom_score_adj:100
[ 3689.324248] Out of memory: Killed process 73395 (firefox-bin) total-vm:3449748kB, anon-rss:89792kB, file-rss:0kB, shmem-rss:5836kB, UID:1000 pgtables:2592kB oom_score_adj:0

Interestingly that crash in event of rclone's doing backup. And firefox-beta is last killed. No mpvpaper in the dmesg.

Before crash input becomes difficulty, high cpu usage, high cpu temperature.

I have mpvpaer pmap here before the crash a minute: https://0x0.st/oAhf.txt

[GhostNaN]

Yeah... I believe you got the wrong guy here. I looked at the pmap and nothing looked out of place. As I said before, you're using less RAM than even me.

If you use htop it should be fairly obvious what is using all of your RAM. Or it could be the fact that you're using /tmp with rclone (you know that's your RAM right?) If not, I wish you luck in finding the source of your memory leak.

[tmpm697]

No, it's something more.

When ram at >90% and I try to kill mpvpaper it drop immediately to less than 10%. This is something.

Memory reported from tools seems not correct.

[tmpm697]

It can be swaywm tries to buffer something that heat up ram.

[tmpm697]

I tried playing a raw 1k video that need even higher cpu/ram to decode its video stream/audio. Playing some youtube videos in multiple firefox-beta tabs Do some ssh sessions as usual. Ram barely reach over 22% and not willing to scale up more than usual.

I think this can due to something with mpvpaper itself or how sway handle drawing hidding mpvpaper window that buffers something in ram.

[tmpm697]

I think I probably blame wrong mpvpaper due to this isssue logged here

[GhostNaN]

That's ok, tracking down memory leaks can be hard. I'm glad you found yours.