kind-cluster

name	kind-cluster
description	"Crea cluster Kind con 1 control-plane y 5 workers (3 GPU, 2 CPU). Valida GPUs con pods nvidia-smi. USA cuando el usuario mencione 'crear cluster kind', 'kind cluster', 'kubernetes gpu', 'test gpu', 'nvidia-smi', o 'kind'."

Kind Cluster - GPU-Enabled Kubernetes Cluster

Objetivo

Crear cluster Kind con:

• 1 control-plane
• 5 workers (3 GPU, 2 CPU)

---

CONFIGURACIÓN

kind-config.yaml

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
name: gpu-cluster
nodes:
- role: control-plane
- role: worker
  labels:
    worker: worker1
    resource: GPU
- role: worker
  labels:
    worker: worker2
    resource: GPU
- role: worker
  labels:
    worker: worker3
    resource: GPU
- role: worker
  labels:
    worker: worker4
    resource: CPU
- role: worker
  labels:
    worker: worker5
    resource: CPU

---

VALIDACIÓN DE GPU

NO USAR: docker exec

USAR: Pod con nvidia-smi

apiVersion: v1
kind: Pod
metadata:
  name: gpu-test-worker1
spec:
  restartPolicy: Never
  nodeSelector:
    worker: worker1
  containers:
  - name: nvidia-smi
    image: nvidia/cuda:12.2.0-base-ubuntu22.04
    command: ["nvidia-smi"]

---

COMANDOS

# Crear cluster
kind create cluster --config=kind-config.yaml

# Verificar nodos
kubectl get nodes --show-labels

# Validar GPU en cada worker
kubectl apply -f test-worker1.yaml
kubectl logs gpu-test-worker1

---

worker_results.md

worker_name	labels	available_gpu	result_by_pod	notes
worker1	worker=worker1, resource=GPU	Yes/No	Success/Failed	-
worker2	worker=worker2, resource=GPU	Yes/No	Success/Failed	-
worker3	worker=worker3, resource=GPU	Yes/No	Success/Failed	-
worker4	worker=worker4, resource=CPU	N/A	Success/Failed	CPU only
worker5	worker=worker5, resource=CPU	N/A	Success/Failed	CPU only