Cluster kind + monitoring

Docker Desktop

Settings → Resources: almeno 4 CPU e 6 GB di RAM. L’LLM gira fuori dalla VM Docker, quindi non va conteggiato qui.

docker desktop status

docker info --format '{{.NCPU}} CPU, {{.MemTotal}} bytes'

Cluster

mkdir -p ~/Downloads/K8S-AI/lab && cd ~/Downloads/K8S-AI/lab

cat > kind-ai-mon.yaml <<'EOF2'
kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
name: ai-mon
nodes:
  - role: control-plane
    extraPortMappings:
      - containerPort: 30090
        hostPort: 9090
      - containerPort: 30093
        hostPort: 9093
EOF2

kind create cluster --config kind-ai-mon.yaml

kubectl cluster-info --context kind-ai-mon

kubectl get nodes -o wide

kube-prometheus-stack

helm repo add prometheus-community https://prometheus-community.github.io/helm-charts

helm repo update

Values per un lab leggero, con Prometheus e Alertmanager esposti su NodePort (mappati su localhost:9090 e localhost:9093 dal config kind):

cat > kps-values.yaml <<'EOF2'
grafana:
  enabled: false
kubeEtcd:
  enabled: false
kubeControllerManager:
  enabled: false
kubeScheduler:
  enabled: false
kubeProxy:
  enabled: false
prometheus:
  service:
    type: NodePort
    nodePort: 30090
  prometheusSpec:
    retention: 12h
    scrapeInterval: 30s
    resources:
      requests: { cpu: 200m, memory: 512Mi }
      limits: { memory: 1Gi }
    serviceMonitorSelectorNilUsesHelmValues: false
    podMonitorSelectorNilUsesHelmValues: false
    ruleSelectorNilUsesHelmValues: false
alertmanager:
  service:
    type: NodePort
    nodePort: 30093
  alertmanagerSpec:
    resources:
      requests: { cpu: 20m, memory: 64Mi }
EOF2

helm upgrade --install kps prometheus-community/kube-prometheus-stack -n monitoring --create-namespace -f kps-values.yaml

kubectl -n monitoring rollout status sts/prometheus-kps-kube-prometheus-stack-prometheus --timeout=5m

kubectl -n monitoring get pods

Verifica dal Mac:

curl -s localhost:9090/-/ready

curl -s localhost:9093/-/ready

curl -s 'localhost:9090/api/v1/query?query=up' | jq '.data.result | length'

Se le NodePort non rispondono, usa i port-forward (terminali separati):

kubectl -n monitoring port-forward svc/kps-kube-prometheus-stack-prometheus 9090

kubectl -n monitoring port-forward svc/kps-kube-prometheus-stack-alertmanager 9093

Grafana (opzionale)

helm upgrade kps prometheus-community/kube-prometheus-stack -n monitoring -f kps-values.yaml --set grafana.enabled=true --set grafana.adminPassword=admin

kubectl -n monitoring port-forward svc/kps-grafana 3000:80

Connettività pod → Ollama sull’host

kubectl run curl --rm -it --restart=Never --image=curlimages/curl -- curl -s http://host.docker.internal:11434/api/version

Se fallisce, vedi Esporre Ollama ai pod di kind.

Stop / start / delete

docker stop ai-mon-control-plane

docker start ai-mon-control-plane

kind delete cluster --name ai-mon