Direct Playback Stutters after 15-45 minutes

[FaptainCrunch7]

Recently, JF has been giving me some issues with playback. JF is direct playing a movie, and after awhile it starts playing in slowmotion. If I back out of the movie and go back in, its fine again for another short amount of time, but then stutters again. I have JF in docker on UNRAID a Ryzen 5600g and 16gb DDR4. When it hangs, CPU is less than 5% with RAM at 20%. I tried with Transcoding enabled and disabled, same issue. Both me and the other user use Android TV and the sever is 10.10.6.

Running a speed test from my AP to the client, I get a consistent 200+ mbps. Same issue also happens to a friend who uses my JF by accessing it on a reverse proxy. Speed test at my router is 350mbps, with my server hardwired directly to it. It never used to do this. Kinda of lost.. any ideas?
Logs from the docker HERE. Logs from JF server HERE

[TheDreadPirate]

Which reverse proxy are you using? Nginx Proxy Manager?

I see that Jellyfin is running in bridge mode. Is the reverse proxy also running with bridge networking? Is the Jellyfin container publishing port 8096?

[FaptainCrunch7]

Which reverse proxy are you using?  Nginx Proxy Manager?

I see that Jellyfin is running in bridge mode.  Is the reverse proxy also running with bridge networking?  Is the Jellyfin container publishing port 8096?

Both JF and NPM are bridge networking and Jellyfin is on port 8096. Everythings been working great for years. I access JF from local lan at home.

[TheDreadPirate]

When you say "jellyfin is on port 8096", does that mean that you are publishing port 8096? I usually see the host's IP in the log when port publishing is used, but I don't see that. Just the bridge IP.

Can you share screenshots of how Jellyfin and NPM are configured in unRAID. And screenshots of the NPM proxy settings for Jellyfin? Censor the domain name.

[FaptainCrunch7]

Added 4 pics for various JF and NPM configs.

[TheDreadPirate]

Turn off "cache assets" in NPM. That can cause a lot of issues with jellyfin, even when direct playing. There is no benefit to it being on and it just consumes a lot of resources.

Unrelated, passing in your GPU via the extra arguments PROBABLY works fine. The "right" way is to click on "Show more settings" and there is an explicit "Devices" field to pass in /dev/dri.

[FaptainCrunch7]

Thanks for the NPM tip! NPM shouldn't be the issue if im just mapping the client while home direct to Jellyfin via local IP while at home though, correct?

[TheDreadPirate]

If you're going directly to port 8096, it would bypass Nginx.

Let's try this. Remove port publishing in Jellyfin. Reconfigure Nginx to proxy Jellyfin's bridge IP (172.16.0.6). Have ALL clients go to the domain address.

To improve the experience for local clients, you should enable NAT loopback if your router supports that. If not, most routers should allow you to add custom DNS entries. You would add a DNS entry to resolve your domain name to your host's LAN IP. I prefer just doing custom DNS entries, in general, so that my domain name still works for LAN clients if the Internet were to go out.

What we're trying to do is reduce the number of context changes that is going on for network traffic. Going from host network to bridge network to host network to bridge network, and then again in reverse.