Poor Performance from Clients

[BeastleeUK]

Configuration:

Jellyfin Version: 10.9.9
Installation Method: Docker (Compose)
OS: Ubuntu 22.04 LTS (Bare Metal)
Reverse Proxy: Traefik v3.1.2

CPU: Intel Xeon E3-1225 v3 @ 4 x 3.2GHz
GPU: Intel i965 GPU
Memory: 20 GB DDR4
HDDS:
  OS and Docker volumes: 2 x 240GB SSD
  Media: 3TB WD Red HDD
Network: 1GBit wired / WiFi6 (clients get 200Mbps+)

Description:
I've been running Jellyfin for over a year but since the 10.9.x releases the performance on my non-web clients has hit a point of being unusable. The web interface is quick as ever and videos, including 4K, play instantly, however the clients on my LG TVs (one running an older version, the other up to date) and my Fire TV Stick (client app v0.17.3) all take up to 2 minutes to start playing and any skipping forward or backward renders it unusable again for a couple of minutes. Playback is DirectPlay to all devices and it affects even 1080p material, host CPU usage sits around 10% and Memory is about 40%, there's nothing obvious in the logs either.  I did try pointing the clients at the IP:port combo to rule out the proxy but that makes no discernible difference.  I've included my Docker and Traefik configs as well as a log capture when I tried to play an episode just now. Please can someone help me avoid going back to Plex?

Logs: https://pastebin.com/a4xa0s4B

Compose File:

Code:

services:
  # Jellyfin - Media Library
  jellyfin:
    hostname: jellyfin
    container_name: jellyfin
    image: jellyfin/jellyfin:latest
    restart: unless-stopped
    profiles:
      - jellyfin
    user: ${PUID}:${PGID}
    group_add:
      - "109"
    network_mode: 'host'
    environment:
      - JELLYFIN_PublishedServerUrl=https://${JELLYFIN_URL} #optional
    volumes:
      - ${DOCKERDIR}/applications/multimedia/jellyfin/config:/config
      - ${MEDIA_CACHEDIR}/jellyfin:/cache
      - type: bind
        source: ${AUTOMEDIADIR}
        target: /automedia
    devices:
      - /dev/dri:/dev/dri

Traefik Config:

Code:

http:

  routers:
    jellyfin-rtr:
      rule: "Host(`jellyfin.mydomain.com`)"
      entrypoints:
        - websecure
      middlewares:
        - chain-no-auth-noindex
      service: jellyfin-svc
      tls:
        certResolver: CloudflareDNS

  services:
    jellyfin-svc:
      loadBalancer:
        servers:
          - url: "http://X.X.X.X:8096"

[TheDreadPirate]

Can you share your full log via pastebin? Censor domain names and WAN IPs.

Also, I see in your docker compose that you are passing in /dev/dri, but your specs don't list a GPU.

[BeastleeUK]

Can you share your full log via pastebin?  Censor domain names and WAN IPs.

Also, I see in your docker compose that you are passing in /dev/dri, but your specs don't list a GPU.

Thanks for taking a look, I've updated my original post to show the i965 GPU and replaced the log block with a link to the PasteBin entry. I tried turning off the transcoding at one point but it had been fine before that with no config changes, it made no difference. Plex is also set up the same way and everything plays instantly on that, including 4K at reasonably high bitrates. There's only one user ever on this at a time too, I have 2 users but we share mine.

[TheDreadPirate]

Ah. I see that this Xeon CPU does have an iGPU in it. But it doesn't have the Quick Sync media engine. But you said it is direct playing, which the logs definitely indicate.

My only suggestion is to try disabling the trakt plugin, restart jellyfin, and try again.

[BeastleeUK]

Ah.  I see that this Xeon CPU does have an iGPU in it.  But it doesn't have the Quick Sync media engine.  But you said it is direct playing, which the logs definitely indicate.

My only suggestion is to try disabling the trakt plugin, restart jellyfin, and try again.

Thanks, I've just tried that and still no difference. It's just weird how it's only the client apps that are affected and not the web interface.

[TheDreadPirate]

Long shot, can you switch your bind mount for /automedia to a plain volume mount?

Code:

- ${AUTOMEDIADIR}:/automedia

[BeastleeUK]

Long shot, can you switch your bind mount for /automedia to a plain volume mount?

Code:

- ${AUTOMEDIADIR}:/automedia

That's actually how it used to be until about 10 minutes before I posted this thread. I was wondering if the volume mount was affecting it.

I'm going to build a new container from scratch and point the mappings, except media, to a new set of folders. I'll see if that changes anything, I'm not bothered if I have to reconfigure anything as Trakt manages my watched times and that's the only thing I care about

[BeastleeUK]

Just to confirm, completely fresh server install on same hardware and with the client app reinstalled too it does the same thing. It took 30s for a 1080p stream to start and any kind of skipping delayed it the same again. Web interface is instant playback.

I have seen other posts around reddit, etc. where the client apps seem to do this but I have now way to confirm if it is the same issue or how to debug it.

EDIT: My client apps are now back to the problem I used to have where they won't connect to Jellyfin. They pick up the published URL but don't connect via the URL or either port. Web clients on the same devices can connect fine.

[TheDreadPirate]

Can you comment out the publishedURL in your docker config? Then connect from a local client directly to your server's LAN IP?

[BeastleeUK]

Can you comment out the publishedURL in your docker config?  Then connect from a local client directly to your server's LAN IP?

Getting weirder. I can connect better over Tailscale VPN when off WiFi (I don't have the ports open on the router). With and without the published URL removed the official client on my phone won't connect even via IP:port. Findroid and web interface will do it no problem so I think it's an Android client issue with the websockets. Really need to be able to view the log on that without ADB.

Thanks,
Lee