Emby Server not releasing CPU process after cancelled HW Transcode

[Happy2Play 9780]

13 minutes ago, softworkz said:

OK, now I to understand what's happening at least:

• When the user changes the bitrate, the client tells the server to stop the playsession
• ffmpeg process is stopped
• Then, the player requests another segment (which shouldn't happen) with the parameters for the stopped playsession
• A new ffmpeg process is created to satisfy the segment request
• The client receives the playlist for the new playsession (with the changes bandwidth params)
• Client requests a segment from that playlist
• Server starts a new ffmpeg process for the that request
• The intermediate process becomes "orphaned"

That's what's happening, I just don't know why... @Luke - any idea?

Is that same thing that is happening here, correct?

 

[softworkz 5066]

@Happy2Play -  Yes, it's the same pattern. Good catch!

[ctaranto 16]

Thanks for looking into this.

Any thoughts on if/when this issue will be addressed?

[Luke 42077]

Hi, it's being looked into, thanks.

[softworkz 5066]

18 hours ago, ctaranto said:

Thanks for looking into this.

Any thoughts on if/when this issue will be addressed?

Intermediate update: the issue that Emby is leaving child processes running on shutdown is fixed for the next beta.

The actual issue is still pending investigation.

[Tau512 15]

bit of a +1 to this.

i'm seeing same the zombie process with symptoms as OP. I discovered it via nvidia-smi tool (Linux server also) when becoming curious how much memory usage per stream-type.

the quadro p400 only has 3 NVENC streams available so aside from the stray processes causing undesirable load, this appears it could actually block future playback of any media type assuming it's transcoding.

source content is 4k

I'm waiting excitedly for the 4.6.x beta to turn final as i've see the changelogs and expectations of the 4.6 release which should bring a lot of useful fixes&features for me hopefully a fix for these zombie processes can be included!