(2023-06-21, 08:25 AM)Venson Wrote:

(2023-06-21, 04:16 AM)natzilla Wrote:

(2023-06-20, 06:14 AM)Venson Wrote:

(2023-06-20, 04:19 AM)joshuaboniface Wrote: Neither error seems related to memory: the first is just a client disconnecting uncleanly, and the second is a playback failure (wrong HWA perhaps). But I'm not sure either of them would really cause such a massive memory leak. For ref, my instance has been up 2 weeks and is only using ~3.3GB

Code:

● jellyfin.service - Jellyfin Media Server
    Loaded: loaded (/etc/systemd/system/jellyfin.service; enabled; vendor preset: enabled)
    Drop-In: /etc/systemd/system/jellyfin.service.d
            └─jellyfin.service.conf
    Active: active (running) since Mon 2023-06-05 21:59:44 EDT; 2 weeks 0 days ago
  Main PID: 476 (jellyfin)
      Tasks: 23 (limit: 19171)
    Memory: 3.3G
        CPU: 2d 2h 38min 29.487s
    CGroup: /system.slice/jellyfin.service
            └─476 /usr/bin/jellyfin --webdir=/usr/share/jellyfin/web --restartpath=/usr/lib/jellyfin/restart.sh --ffmpeg=/usr/local/bin/ffmpeg

---

It would also help to have more details on the specifics of your setup: what version, what OS, package format, etc.

Looking at the logfile he has a lot of errored websocket connections, maybe there is actually a memory leak there?
On the other hand the 
Webhook Item Added Notifier
message comes every few sec which could be an issue?

@natzilla i would recommend disabling all plugins and see if that helps as a troubleshooting step.

Making that change it did take longer for the ram to fill up. It does at least appear like it's going down when no activity is present on it, but to still get as high as 16G? that still seems much to me. I have included a new batch of logs on this reply.

Are you sure that graph only shows jellyfin memory and not "overall" system memory?

(2023-06-21, 04:31 PM)natzilla Wrote: Yes I am sure. That graph is just for jellyfin only.

(2023-06-21, 05:52 PM)joshuaboniface Wrote: Good catch @Venson, yea @natzilla your actual Jellyfin memory usage is only 2.9G as reported by your systemctl output, and that seems normal. Something else is using up all your memory.

---

(2023-06-21, 04:31 PM)natzilla Wrote: Yes I am sure. That graph is just for jellyfin only.

For your Jellyfin *system* but not the process. The process itself is only using what systemctl reports.

Try posting your entire ps aux output, or htop sorted by %Mem, that will show what's really using it. If it is Jellyfin then we learned something new about systemctl's accuracy, but I suspect it's not.

(2023-06-21, 08:09 PM)joshuaboniface Wrote: Looking through the entries, Jellyin itself is using 17% of the memory, which if it's 16G is 2.7GB so pretty close to what systemctl said. FFMpeg is also using another 1.1%, and since it's a child process of Jellyfin, it would be counted, so we're basically at the ~3GB that systemctl reported:

Code:

USER        PID %CPU %MEM    VSZ  RSS TTY      STAT START  TIME COMMAND
[///]
jellyfin    2204  3.2 17.7 7034120 2850508 ?    Ssl  Jun20  51:08 /usr/bin/jellyfin --webdir=/usr/share/jellyfin/web --restartpath=/usr/lib/jellyfin/restart.sh --ffmpeg=/usr/lib/jellyfin-ffmpeg/ffmpeg
jellyfin  15988  2.9  1.1 9498676 191252 ?      Sl  19:16  0:23 /usr/lib/jellyfin-ffmpeg/ffmpeg -analyzeduration 50000000 -init_hw_device cuda=cu:0 -filter_hw_device cu -hwaccel cuda -hwaccel_output_format cuda -threads 1 -a

However, nothing else in the output seems to be using any more RAM than this.

Looking back at your graph, all it says is "Memory Usage". What monitoring tool is this? It's very likely that what you're seeing is not Jellyfin - or anything - actually using all the RAM, but Linux's page cache being put to use. https://www.linuxatemyram.com/ This is super common to see in more Linux-naive monitoring tools that treat anything using memory as "used" even though that memory is actually free to applications.

You can check this yourself with free -m to see buff/cache, available and free; free will, after some time, start to get very low, but buff/cache increases along with available. buff/cache is memory used by the page cache i.e. caching files on the filesystem in memory for faster access, while available is the memory in the cache that is actually available for appliations to use.

Graphing/monitoring Linux memory usage should be differentiating between these values so that one could see the actual application memory usage versus page cache usage, but your graph doesn't seem to be, so I suspect you're not seeing the whole picture there and thus thinking that Jellyfin is "using all the RAM" when in reality it's only using ~19% of it.

Really, things only become a problem when you actually start seeing swap space being used. If swap isn't being used, even sitting at "100%" memory usage is fine as it's all page cache. Once things start swapping, or you start getting Out Of Memory errors in your syslogs (or dmesg), then you are truly out of memory.

Hopefully that helps!

(2023-06-21, 11:58 PM)joshuaboniface Wrote: Aah knowing that it's ProxMox definitely explains it. Because of how the KVM virtualization in ProxMox works, from ProxMox's perspective, the RAM is indeed "used" by the VM even though it's just page cache data inside the VM, because the hypervisor can't hand that RAM out to other VMs. Unlike inside the VM itself where that memory could easily be used by another application.

Best thing to do would just be to shrink the VM. If Jellyfin isn't using 16GB (and it isn't, not even close), the VM doesn't need to be 16GB and thus it won't steal 16GB of RAM from the hypervisor. Based on your usage a 4GB VM is probably plenty.

(2023-06-24, 07:50 AM)joshuaboniface Wrote: OK there's definitely something wrong there, it should not be consuming that much RAM. It's definitely weird that ps aux is reporting so much less than systemctl as well. Unless someone beats me to it I'll check out the logs soon.

(2023-06-26, 03:19 PM)natzilla Wrote:

(2023-06-24, 07:50 AM)joshuaboniface Wrote: OK there's definitely something wrong there, it should not be consuming that much RAM. It's definitely weird that ps aux is reporting so much less than systemctl as well. Unless someone beats me to it I'll check out the logs soon.

I want to add it's possible it might have something to do with transcode, but I'm unsure. Least from what behavior I saw over the weekend.