Brand new here. Cant get a single pod to start

[visual idol]

Always getting the error below. 150GB disk space, 150GB container space (tried different variations). pytorch cuda 12.81 template with modified disk spaces

create container runpod/pytorch:1.0.2-cu1281-torch280-ubuntu2404
1.0.2-cu1281-torch280-ubuntu2404 Pulling from runpod/pytorch
Digest: sha256:0a360022e8de4375af99430f84e8b38951acc397252163a37ceac7204d01be35
Status: Image is up to date for runpod/pytorch:1.0.2-cu1281-torch280-ubuntu2404
error creating container: container: create: container create: Error response from daemon: failed to mkdir /var/lib/docker/231072.231072/volumes/sffdhqn77vh3b3/_data/.cache: mkdir /var/lib/docker/231072.231072/volumes/sffdhqn77vh3b3/_data/.cache: no space left on device

[floral niche]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[summer spade]

this has definitely nothing to do with your pod, it's basically "Docker host RunPod put you on is full". Fastest solution would be to terminate this pod and create a new one (even better: create a new one in a different location). Also you dont need 150gb for a container disk, 30gb should be more than enough

[cosmic marsh]

Some people do need them that big, you don't know his workload

I'm a 125GB container disk 0 volume disk user for example

[summer spade]

you are right, I dont. I just assume that most people like to persist their data. For example I keep root's home, all python venvs, all application settings (comfy, jupyter, kohya, etc ..), models and outputs on /workspace volume so it persists all my settings after reboot. If he does not need persistence for his data, there is very little need to use pod and I'd go serverless instead.

@cosmic marsh out of curiosity - so what is your usecase that you need 1--+gb for container and no permanent storage?

[cosmic marsh]

Running LLM's

The KoboldCpp template downloads them faster than the persistent storage can load them

[summer spade]

ok, makes sense.

[cosmic marsh]

And KoboldCpp bundled UI is designed around downloading your saves, there is no server side saving by default

So we actually don't have changing data of any kind

Gives me full freedom to switch to any GPU in any datacenter at any time or delete them when I am done with them 😄

It also works on every single kind of runpod instance so full flexibility

[cosmic marsh]

Btw for your specific use case I made a template at https://henk.tech/ccrunpod it also automatically puts the user homefolder on the volume/network storage

Didn't know other people did that to outside of my template, its not the easiest trick for people to do 😄

[summer spade]

agree. I've tried like 10 community templates for stable diffusion and was really annoyed with setting dark mode on jupyter every single time, then doing "source venvs/..." on 4 terminals, etc. So I did my own template too :D https://console.runpod.io/deploy?template=gg1utaykxa&ref=o3idfm0n

[cosmic marsh]

How'd you discover how to do it?

Or were you like me and just experimented with sysadmin knowledge haha

[summer spade]

well, been playing with linux/unix systems for 30 years now :)

[cosmic marsh]

Looks like its not actually on the volume storage

Or waitn o it is

But you put it in a subfolder

Very clean

I chose to have the home folder be directly on /workspace