ffmpeg with nice/ionice #1192
ffmpeg with nice/ionice #1192
Conversation
| @@ -25,6 +25,7 @@ func MuxTStoMP4(tsInputFile, mp4OutputFile string) ([]string, error) { | |||
| // transmux the .ts file into a standalone MP4 file | |||
| ffmpegErr := bytes.Buffer{} | |||
| err := ffmpeg.Input(tsInputFile). | |||
| SetFfmpegPath("/path/to/ffmpeg-nice.sh"). | |||
There was a problem hiding this comment.
| @@ -81,7 +82,7 @@ func MuxTStoFMP4(fmp4ManifestOutputFile string, inputs ...string) error { | |||
|
|
|||
| timeout, cancel := context.WithTimeout(context.Background(), 10*time.Minute) | |||
| defer cancel() | |||
| cmd := exec.CommandContext(timeout, "ffmpeg", args...) | |||
| cmd := exec.CommandContext(timeout, "/path/to/ffmpeg-nice.sh", args...) | |||
There was a problem hiding this comment.
This is pseudo-code or just a starting point for discussion. I'm not even sure if these two places are the correct ones for VOD processing
There was a problem hiding this comment.
This is just the MP4 generation step. There's ffmpeg usage in different places for for segmenting, thumbnailing, clipping, etc. Just grep-ing ffmpeg or CommandContext | grep ffmpeg should get you most of the instances.
There was a problem hiding this comment.
So we're not fully sure yet but it looks like some ffmpeg for big VODs are congesting both disk writes on datapacket servers and generate big egress spikes resulting in entire server slowdown and k3s restarts.
We're testing with Rafał if moving VOD processing onto just one of the nodes in the cluster results in less or zero restarts of the rest of the nodes.
|
What problem are we trying to solve? nice-ing to such a low priority may make the full transcode cycle much slower. Did this change make some problems go away? |
pwilczynskiclearcode
force-pushed
the
ionice
branch
from
33e8784 to
752e8a1
Compare
No description provided.