FlexibleToast

Btw, at least for Ollama specifically I did have to leave the --gpus=all otherwise it never saw the GPU. I think that is on a per container basis though because it seems like some others did not need it.

I removed the --runtime=nvidia from my Ollama's "Extra Parameters" and it seems to work just fine. It's even still using the GPU. I did leave the --gpus=all but, I don't know if that is needed.

I think you're having the same issues I'm having. Is that one container using the Nvidia runtime? All of my Nvidia runtime containers fail to start with that same error. I made sure I have the latest Nvidia driver.

Or at least semi random. They seem to happen occasionally when I'm running AI inference workloads like Ollama. I'm suspecting it's heat related, but this time the room was relatively cool. Like the room was cooler than it usually is. Is there any insight in this diagnostic as to what caused the last reboot? It just happened right before this diagnostic was pulled.kerbol-next-diagnostics-20250920-2100.zip

That's not a bad idea. It will take a little bit of planning now that I'm using the Nextcloud AIO container that uses docker storage. Granted, that's what the borg backups are for I suppose. Thanks for the help on this. Glad it was as simple as just a drive (not even one in the array) causing issues. Since that drive is just there as a local backup, I'm wiping it now and starting fresh.

I agree with your assessment. However, I used to use the docker image in the past (it is the default after all). That image got corrupted and while it was relatively easy to recover from because all the container settings were saved by Unraid, it was annoying to have to do. That's when I switched to the folder method. I also, didn't realize at the time it would use the ZFS driver and create all those snapshots. That's the "I realized that my sanoid/syncoid settings where messed up". On top of docker creating lots of snapshots, sanoid/syncoid where also creating a bunch of snapshots and causing general havoc. Now I have system/docker excluded from those tools.

That seemed to be it. I rebooted the server and it was back up and array started in 10 minutes... Guess I need to come up with a different way to store local zfs snapshots.

Oh, the settings finally opened. I've turned off automount. I'll try rebooting later tonight when I'm not about to watch Plex.

I think you're on to something, because I can't. I can't umount it because target is busy. If I try to lsof the mount I get spammed with lsof: no pwd entry for UID 1001 and various other UIDs. I realized that my sanoid/syncoid settings where messed up and where essentially stuck running. In the UI I've tried pressing the cog next to the device so I can change the automount setting, but it never opens...

Yes, that diag is directly after that boot. I can tell you that my array wasn't actually started until about 10:36, that's when docker finally came online and my containers came back up. It sounds like the disks and pools are working as expected, but for some reason the array is taking a really long time to form/start.

This is getting brutal. Something caused the system to hang last night around 00:30. I had to reboot it and it took 1:40 for the array to come up. Nearly 2 hours, that's wild. I'm thinking of going back to Proxmox as my homeserver, virtualizing unraid, moving all/most of my docker stuff off to a different vm, and let unraid just be storage. kerbol-next-diagnostics-20250910-1040.zip

It's weird that I didn't get the notifications for updates. I'm on v7.1.2 now and it still doesn't work.

I swear I had this working in the past... I'm positive at one point Plex was using hardware transcode just fine and now it seems nothing can use it. Relevant specs/info: Intel i9-12900KS Unraid 7.0.1 Intel-GPU-TOP app installed Relevant containers have `--device=/dev/dri:/dev/dri` as extra params I suppose I first noticed with tdarr that QSV wasn't working. That would have been around March 2nd. I didn't think much of it, but now after setting up Jellyfin I really started to look into things. I installed intel-gpu-tools in the Jellyfin container and `intel_gpu_top` would error and not find any GPUs. I then looked in `/dev/dri` and renderD128 is there along with card0. I tried the same thing on the Unraid host and it didn't work. I even tried modprobe i915: root@Kerbol-Next:~# modprobe i915 root@Kerbol-Next:~# intel_gpu_top No device filter specified and no discrete/integrated i915 devices found root@Kerbol-Next:~# ls /dev/dri by-path/ card0 renderD128 I'm not sure what else to look for and test? kerbol-next-diagnostics-20250508-0930.zip

Correct, but that docker.img is on ZFS.

I haven't. But, we're talking a ZFS pool here. ZFS won't just barf out corruption like that. That shouldn't be the problem, but I should check anyway.