Nimendra's Notes 🪴

Metrics Server

2 min read

kubernetes kubernetes/cka

Logging & Monitoring
Observability

last9.io/blog/kubernetes-metrics-server/
Scalable and efficient source of container resource metrics for Kubernetes built-in autoscaling pipelines.

Core Idea

Metrics Server is the lightweight, in-memory source of CPU and memory numbers behind kubectl top and the HPA; it is not a monitoring stack.

  • It collects key usage data and serves it to the API server through the Metrics API: short-term, in-memory only, no long-term storage or complex queries.
  • Division of labor: Metrics Server for real-time usage and autoscaling, kube-state-metrics for object metadata, Prometheus for long-term storage, queries, and alerting.
  • Setup notes: --kubelet-insecure-tls works around certificate problems, K3s ships it by default, and the HPA example holds CPU near 50%.

Intro

The Kubernetes Metrics Server collects key data like CPU and memory usage and shares it with the Kubernetes API server through the Metrics API.

  • It’s a lightweight tool with short-term, in-memory metrics storage.
  • It’s not built for storing metrics long-term or handling complex queries.
  • It plays a critical role in autoscaling your workloads in Kubernetes.

Comparing with Other Solutions

  • The Metrics Server is all about tracking real-time resource usage (like CPU and memory) for pods and nodes.which is essential for autoscaling.
  • Kube State Metrics, on the other hand, focuses on the metadata of Kubernetes objects—things like the status of deployments, nodes, and pods.
  • Prometheus, meanwhile, goes beyond that. It offers long-term storage, more advanced querying options, and alerting features, making it a go-to for more complex monitoring needs.

Setting Up the Metrics Server

If you’re running into TLS certificate problems, try using the --kubelet-insecure-tls flag.

apiVersion: v1
kind: ConfigMap
metadata:
  name: metrics-server-config
  namespace: kube-system
data:
  args:
    - --kubelet-insecure-tls
    - --kubelet-preferred-address-types=InternalIP

🌟 In K3s Metrics server available by default.

Accessing metrics server

kubectl top nodes
kubectl top pods

HPA

In this setup, the HPA adjusts the number of pods in the php-apache deployment to keep CPU utilization around 50%.

apiVersion: autoscaling/v2beta1
kind: HorizontalPodAutoscaler
metadata:
  name: php-apache
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: php-apache
  minReplicas: 1
  maxReplicas: 10
  metrics:
    - type: Resource
      resource:
        name: cpu
        targetAverageUtilization: 50

Backlinks

Graph View