[Docs] update docs for setting up gpu metrics on nebius

[rohansonecha]

The 1-click Deploy Prometheus Operator Application on Nebius no longer works because it relies on pulling Bitnami images that are not longer being served publicly on DockerHub. This PR updates the instructions in our docs for setting up gpu metrics on nebius to use a command to manually deploy the prometheus-community/kube-prometheus-stack chart instead.

I used this command to set up GPU metrics on a test api server and verified that using this command instead of the Nebius 1-click Deploy Prometheus Operator Application and then proceeding to follow the rest of the existing doc worked as expected.

Tested (run the relevant ones):

• Code formatting: install pre-commit (auto-check on commit) or bash format.sh
• Any manual or new tests for this PR (please specify below)
• All smoke tests: /smoke-test (CI) or pytest tests/test_smoke.py (local)
• Relevant individual tests: /smoke-test -k test_name (CI) or pytest tests/test_smoke.py::test_name (local)
• Backward compatibility: /quicktest-core (CI) or pytest tests/smoke_tests/test_backward_compat.py (local)

[rohansonecha]

/build-docs

[github-actions]

✅ ReadTheDocs build triggered for branch update-nebius-docs

The documentation will be available at: https://docs.skypilot.co/en/update-nebius-docs/

[DanielZhangQD]

Thanks! @rohansonecha

[DanielZhangQD]

Prometheus is required for the external Nebius cluster, the cluster is assumed to be an external cluster in this doc, right? For in-cluster configurations, we may need to include them in the doc here?

[rohansonecha]

This is a great catch, I will update the command. Also, it seems that Nebius is working on a fix for their Prometheus Operator application. I wonder if we should hold off on merging this PR until that work is done? Or just update our docs with the change in this PR to use the prometheus community chart to be future proof.

[rohansonecha]

Updated the command to enable prometheus @DanielZhangQD

[rohansonecha]

I also updated the in-cluster doc.

[DanielZhangQD]

This is not required, users can also deploy Prom without operator.

[rohansonecha]

Addressed

[DanielZhangQD]

Thanks! @rohansonecha

[DanielZhangQD]

I think we can keep the application installation method, users need to install the device plugin application on the console anyway. Or add these commands as a backup? Tell the users that if this way fails, they can install it manually.

[rohansonecha]

Yep, I think it makes sense to add the commands as a backup. It is easier to install the Prometheus Operator via the Nebius console, but I think there is still value in including the instructions for how to install the operator manually using the community chart in case the Nebius 1-click deploy is down.

[DanielZhangQD]

What about we separate the command into the Check the node exporter setup and Prometheus setup sections as example commands.

[rohansonecha]

I added a section before Check the node exporter setup and Prometheus setup with instructions for deploying the prometheus operator and node exporter.

[DanielZhangQD]

The Prometheus Operator is necessary for the DCGM-Exporter to start properly. This is not true, these two components are independent. Let's just add the command as example commands as described in the comment for L38.

[rohansonecha]

Removed this line

[rohansonecha]

@DanielZhangQD this PR is ready for a final review. I ended up removing the in-cluster instructions from this page, as I felt it was adding additional complexity/confusion to the docs which are already quite clear. Please take a look when you get a chance!