Skip to main content

​
What to monitor

ComponentWhat to monitor
API Server (kube-apiserver)Request rates, error rates, and latency
Scheduler (kube-scheduler)Scheduling latency and errors
Controller Manager (kube-controller-manager)Status of various controllers
  • The number of nodes in the cluster
  • CPU and memory usage — ensure nodes are not over or underutilized
  • Disk usage — monitor disk space and I/O operations
  • Network — monitor bandwidth and errors
  • The number of pods in the cluster
  • Resource usage — CPU, memory, and disk usage per pod and container
  • Pod status — Running, Pending, Failed, and other states
  • Container logs — errors and warnings
  • Health checks — results of liveness and readiness probes
  • Application metrics — custom metrics such as request rates and error rates

​
Metrics Server

You can use other metrics solutions such as Prometheus, Datadog, or Dynatrace in place of Metrics Server.
Metrics Server collects resource metrics (CPU and memory usage) from each node’s kubelet and aggregates them. It then delivers the data to the Kubernetes API server (kube-apiserver) for autoscaling purposes.
  • Horizontal Pod Autoscaler (HPA) uses this data to automatically adjust the number of pod replicas in a deployment.

​
How metrics are collected

Each kubelet contains a component called cAdvisor (Container Advisor) that collects resource usage data from running pods and exposes it through the kubelet API. Metrics Server reads from this API to gather cluster-wide resource data.

​
Commands

# View performance metrics of nodes
# Requires Metrics Server to be correctly configured and running
kubectl top node

# View performance metrics of pods
kubectl top pod
The kubectl top commands require Metrics Server to be correctly configured and running in the cluster.

Build docs developers (and LLMs) love

Get started for freeTalk to us