OOM (Out Of Memory) kills for ffmpeg.

[tuxx]

I've been using mediacms locally with a virtual machine of 4 threads, 4gb ram, and an ssd of 100gb (According to These docs ). I'm basically the only one adding videos, and there are maybe a maximum of 5 people that watch 1-3 videos per day.

I see alot of OOM kills when i add alot of videos, which keeps the encoding processes in limbo in the admin interface.

Wondering if there is any way to keep the encoding from starting when the machine is using alot of performance for other encoding jobs. Maybe set a maximum number of threads for the ffmpeg encodes or something?

Would love to hear your thoughts.

Server specs

# lscpu
Architecture:            x86_64
CPU(s):                  4
  On-line CPU(s) list:   0-3
Vendor ID:               GenuineIntel
  BIOS Vendor ID:        QEMU
  Model name:            Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz

# free -m
               total        used        free      shared  buff/cache   available
Mem:            3914        2542         950          27         681        1372
Swap:              0           0           0

# df -h /
Filesystem      Size  Used Avail Use% Mounted on
/dev/vda1        99G   43G   53G  45% /

Logs

# journalctl -b 1 | grep "Out of memory"

<snipped alot of lines for readability>

Mar 03 03:56:14 mediacms kernel: Out of memory: Killed process 620392 (ffmpeg) total-vm:1679456kB, anon-rss:369960kB, file-rss:0kB, shmem-rss:0kB, UID:33 pgtables:1284kB oom_score_adj:0
Mar 04 23:18:27 mediacms kernel: Out of memory: Killed process 981087 (ffmpeg) total-vm:1733016kB, anon-rss:468432kB, file-rss:0kB, shmem-rss:0kB, UID:33 pgtables:1452kB oom_score_adj:0
Mar 04 23:18:35 mediacms kernel: Out of memory: Killed process 981344 (ffmpeg) total-vm:1733004kB, anon-rss:465700kB, file-rss:0kB, shmem-rss:0kB, UID:33 pgtables:1440kB oom_score_adj:0

# journalctl -b 1 | grep "Out of memory" | wc -l
609

[tuxx]

Fixed this now by increasing memory for the virtual machine, and in my youtube2mediacms checking when encoding is done for an upload before uploading a new video.

[mgogoulos]

Hi, thanks for reporting. I remember in a number of cases that I've uploaded a big number of videos in a single server, in one case it would run ffmpeg and transcode for 2-3 weeks before completing all videos/profiles. The idea is that a single task runs per cpu core (a task can be ffmpeg on a single video for a single profile, or ffmpeg on a single chunk of a video for a single profilke), so that all cpus are occupied. I'm not sure sure what the condition could be that it ended in OOM, eg could it be running other programs as well?

If this happens again, do check the number of ffmpeg tasks running (eg ps aux|grep ffmpeg|wc -l or similar)

Thanks!

[tuxx]

I will replicate the behaviour i've seen to see what is going on.

Currently my youtube2mediacms seems to handle this perfectly by waiting on every encoding profile for an uploaded video to complete before uploading the next. So its not really an issue anymore (for me atleast) 😂