Playback slow to start, randomly stopping

[SierraThePotato]

Hello,

I am having trouble finding out what is the problem with my Jellyfin install. Sometimes it works flawlessly, other times it takes minutes to start the stream, and even then, it stops again after a few seconds. Hardware should be plenty, and when checking utilization nothing is maxing out.

Setup:
Install method: linuxserver.io docker container, in a Debian VM running in Proxmox.
CPU: i5-8500T, 4 cores allocated to the VM
RAM: 20GB allocated to the VM
OS disk: NVME SSD, this is where the docker container is installed, config and cache are here
Media disk: 2TB Seagate HDD, this is where the media files are stored
External traffic is handled by a Caddy reverse proxy, but the issue persists when I use the direct IP over Tailscale. It's been a while since I used it on the local network, so I don't know about that

Hardware transcoding is enabled, although I'm not sure if it is working.
I can provide the docker compose if that helps.

Here are the log files:

  log_20250117.txt (Size: 54.97 KB / Downloads: 47)

  FFmpeg.DirectStream-2025-01-17_11-30-06_5ac9c1f4415b9250c835e09fcd7c9224_91fc7659.txt (Size: 65.78 KB / Downloads: 49)

[TheDreadPirate]

Are you running out of disk space? How much free space is on the drive that you have /config mounted to?

[SierraThePotato]

I was, until a few day ago, but noticed it and solved the issue. It is a 100GB partition and it has 64GB of free space. The issue didn't go away unfortunately. I mostly watch 720p/1080p content, so that should be plenty.

[TheDreadPirate]

Try playing videos from the local network. Just to eliminate tailscale or caddy issues as a factor.

[SierraThePotato]

Will do tomorrow. But what would cause both Caddy and Tailscale to have this issue?

[TheDreadPirate]

No idea. Just trying to eliminate factors since we don't have much information to work with in your logs.

Can you also try disabling tailscale on the server and accessing Jellyfin through Caddy?

[SierraThePotato]

I installed Tailscale recently, and this issue was present before that. Also, on Tailscale I use the IP address, and with Caddy I use a URL, so I can go around Tailscale, directly to Caddy, but the issue is the same.
Should I enable debug logging and check if anything shows up there?

[TheDreadPirate]

I've rarely found debug logging to be helpful. In addition to being hugely verbose as to hide any useful logs in hundreds of thousands of useless logs.

A couple things you can try.

1. Switch to host networking
2. Spin up a test container, but with the official jellyfin/jellyfin image instead of the linuxserver image.

Regard #2, I've seen weird bugs like this that only happen with the linuxserver image but not the official image.

[SierraThePotato]

I've been testing it today on the local network, but going through Caddy, using the URL, so essentially it's going out my router, and coming right back in (at least that's what the tracert command tells me). It was flawless, started immediately, never stopped during playback.
So does this mean the issue is between my LAN where the server is and the LAN where the client is?
Internet speed should not be a problem. The LAN with the server has 95 Mbps upload, and the LAN with the client has 200 Mbps download.
I also switched the networking mode to host, so I will try from the other LAN to see if it helped. If not, I will try switching to the official docker image.