RunPod•3w ago

SSH key generation issue with `runpodctl config`

I'm receiving a new error as of today when I run runpodctl config --apiKey $RUNPOD_API_KEY

Error: failed to update SSH key in the cloud: failed to get SSH keys from the cloud: API error: Unauthorized

Error: failed to update SSH key in the cloud: failed to get SSH keys from the cloud: API error: Unauthorized

I wasn't getting this error before today. I tried generating a new API key (with full perms) just in case there was an issue with it (even though I could use it successfully with a GraphQL request). Still fails. I added a print statement before this to confirm that $RUNPOD_API_KEY has the correct value.

17 Replies

Dj•3w ago

I think I know what's wrong 😅 runpodctl honestly needs a lot of love. Weirdly enough this might have to do with your operating system, but I'm not good enough at Go to understand exactly where the issue comes from.

Dj•3w ago

https://github.com/runpod/runpodctl/issues/160#issuecomment-2348458698 Take a look here

GitHub

runpodctl config command fails on linux · Issue #160 · runpod/run...

Hello, I am facing an issue configuring runpodctl on Ubuntu. When I run the command: runpodctl config --apiKey <my_api_key> I receive the following response: Runpod config file not found, ple...

billpressOP•3w ago

I did get that error message, but it went away when I created an empty ~/.runpod/config.toml (by touching it). Then, it worked fine until today, when I started getting this new ssh-related error. Oh, I see that ssh error further down in the comments. So... what do I do to work around this? fwiw, I'm only doing this so I can get the pod's ID from within the docker container:

export POD_ID=$(runpodctl get pod | awk -v name="$POD_NAME" -F$'\t' '$2==name {print $1; exit}')

export POD_ID=$(runpodctl get pod | awk -v name="$POD_NAME" -F$'\t' '$2==name {print $1; exit}')

Dj•3w ago

echo $RUNPOD_POD_ID 👀

billpressOP•3w ago

Oh! No way Will that work?

Dj•3w ago

I have a meeting in the morning to figure out who owns repos like this one so I can figure out what we can do. But at least you're good in the meantime.

Dj•3w ago

ye here's a list https://docs.runpod.io/pods/references/environment-variables

Pod environment variables | RunPod Documentation

Configure and manage your pods with these essential environment variables, including pod ID, API key, host name, GPU and CPU count, public IP, SSH port, data center ID, volume ID, CUDA version, current working directory, PyTorch version, and public SSH key.

Dj•3w ago

or just do printenv from inside a pod and skim for the RUNPOD_ ones

billpressOP•3w ago

Oh wow, ok. That is so much easier. Thank you!

Dj•3w ago

Dj•3w ago

billpressOP•3w ago

Amazing. Thank you! @Dj , I still want to run these commands at the end of my container's CMD:

    # Send the output.log file to the pod when the command ends
    runpodctl send --code "$SEND_CODE" /workspace && \
    # Stop this pod so it isn't restarted
    runpodctl stop pod "$RUNPOD_POD_ID"

    # Send the output.log file to the pod when the command ends
    runpodctl send --code "$SEND_CODE" /workspace && \
    # Stop this pod so it isn't restarted
    runpodctl stop pod "$RUNPOD_POD_ID"

I found the REST API call for stopping the pod. Is there a REST API call you recommend for sending the /workspace directory using the $SEND_CODE (which was generated by the client creating the container, so it can call receive on its end).

Dj•3w ago

The send command is powered by croc. We have S3 APIs for accessing that workspace directory like an S3 bucket coming out within the next 2 weeks* *Not my release so please don't be surprised if it doesn't happen but it should

billpressOP•3w ago

@Dj Is there a plan to fix the ssh keys issue with runpodctl config? That would be much simpler from a client perspective, so I can just simply run runpodctl receive <key> in my workflow and it will all just work.

Dj•3w ago

Yeah of course, I just don't write Go so I can't confidently Go fix it myself :frowning3: Past a certain point the issue outlevels my skillset Unless there's an open PR right now that fixes it - update: there isn't

billpressOP•3w ago

Totally get it. No expectations of you, personally. More wondering about eng is prioritizing as a team. Sounds like it's not unblocking linux users to use runpodctl. So here's my use case: I'm training models and writing logs to local disk (as well as some trainer models/binaries). Once the training script completes, I need to copy the logs and binaries back to the workflow/user so I can shutdown the pod. I don't want to use a network volume, as I want to be able to grab whatever pod is available of a given type (e.g., H200) anywhere in the country (whereas using a network mounted volume limits the associated pods to the same region). Do you have suggestions for another way to do this while runpodctl send/receive is broken?

Dj•3w ago

Immediate thoughts are to use sftp and just do a push from the pod before it dies - but that would require you to have an SSH key setup which is part of the issue. From the Pod you could also push to an S3 bucket? Does your setup allow you to use runpodctl send/receive and pass the code? You could use magic-wormhole, or just use croc directly since that's all runpodctl send/receive is..

Gaming

Programming

SSH key generation issue with `runpodctl config`

Did you find this page helpful?