skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

orchestration-setup

Audit, prepare, and deploy PAIDF Orchestration on Kubernetes GPU clusters.

3.2k|370|Updated Feb 25, 2026
Search Tags:#audit#kubernetes#Airflow# orchestration#NVIDIA#helm deployment#gpu cluster
One-click install
npx skills add https://github.com/NVIDIA/skills --skill orchestration-setup
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: orchestration-setup
Source: https://github.com/NVIDIA/skills/tree/main/skills/paidf-orchestration-setup
Command: npx skills add https://github.com/NVIDIA/skills --skill orchestration-setup

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

Setting up a PAIDF Orchestration environment requires auditing GPU capacity, resolving kubeconfig access, validating secrets, deploying an Airflow controller via Helm, and verifying controller readiness — a multi-step process where mistakes like pre-creating Helm-managed secrets or using unreachable kubeconfig server addresses cause hard-to-diagnose failures.

Core Features & Use Cases

  • Read-only cluster audit: Runs scripts/remote_k8s.py audit locally or over SSH to report GPU capacity, namespace, registry secret, and storage class readiness as structured JSON without mutating the cluster.
  • Guided controller deployment: Walks through make setup (secrets validation and Helm values generation) and make install sdg-controller, with explicit approval gates, S3 connectivity pre-flight, and post-install verification against live cluster state.
  • Topology and service-mode selection: Helps choose controller placement and external versus in-cluster model services independently, recommending external endpoints for single-GPU H100 nodes.
  • Use Case: A user with one H100 host running k3s asks to set up SDG Workflow; the skill audits the cluster over SSH, recommends external model endpoints, validates secrets.env, deploys the Airflow controller, and verifies pools, DAGs, and connections before reporting readiness.

Quick Start

Ask the agent to audit your GPU cluster for PAIDF Orchestration and provide the path to your kubeconfig when prompted.

Frequently Asked Questions about orchestration-setup

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I deploy the SDG controller on a Kubernetes GPU cluster?▼

Run `make setup` first to validate secrets.env and generate Helm values, then run `make install sdg-controller` to upload DAGs to S3 and install the Helm release into the sdg-workflow namespace. Verify afterward with helm list and kubectl get pods against the live cluster state.

How do I audit a remote GPU cluster over SSH before setup?▼

Run `python scripts/remote_k8s.py audit --ssh-target user@host --kubectl-command "k3s kubectl" --service-mode external --json`. The audit is read-only and reports ready GPUs, namespace, registry secret, and storage classes as JSON without changing the cluster.

Can I run PAIDF Orchestration on a single H100 GPU node?▼

Yes, a single-GPU H100 node is supported when using external model endpoints, since internal mode requires at least three to four allocatable GPUs. The Airflow controller and augmentation worker can still run in that node's cluster.

Why does helm install fail with an ngc-docker-registry-secret ownership error?▼

The error occurs when the secret was pre-created manually without Helm ownership metadata, so the chart cannot adopt it. Delete the secret with kubectl and let `make install sdg-controller` recreate it during installation.

Does the skill support a Docker-only or plain SSH host without Kubernetes?▼

No, a Docker-only or SSH-only host is not a supported backend. You must first install a Kubernetes distribution and the NVIDIA device plugin so the cluster exposes nvidia.com/gpu resources, then rerun the audit.

How do I update Airflow DAGs without reinstalling the Helm chart?▼

Run `make sync-dag` to push updated DAG files to S3. The dag-synchronizer sidecar pulls changes at the configured interval (default 30 seconds), so no helm upgrade is needed for DAG-only updates.

Related Skills

Explore More Software Engineering Skills→
nebiusnebius
17

groot

Manage NVIDIA GR00T model deployment, finetuning, and inference on Nebius clusters.

Official
Advanced
rory-datarory-data

airflow

Automate Airflow DAG operations and health checks via CLI commands.

Community
Advanced
SerendipityOneIncSerendipityOneInc

k8s-management

Inspect Kubernetes cluster status, logs, and resources via read-only kubectl commands.

Official
Intermediate
NousResearchNousResearch
239.1k

modal

Deploy and run ML workloads on Modal's serverless GPU cloud platform.

Official
Intermediate