• How to monitor resource consumption?
  • Kubernetes currently does not come with a monitoring solution.
  • There are solutions available for this, those being:
    • Metrics Server
    • Prometheus
    • Elastic Stack
    • Datadog
    • Dynatrace
  • Heapster was an original project that allowed a lot of monitoring and analysis of a cluster for Kubernetes.
    • Heapster is now deprecated and Metrics Server is a slimmed down version of that.
  • You can have one Metrics Server per cluster.
    • It retrieves metrics from each Kubernetes node and pod, gathers data and aggregates it.
    • This is all done in memory.
      • Not on disk.
      • Cannot see historical performance data.
  • How are the metrics generated on each node?
    • Kubernetes has a kubelet per node.
      • Response for receiving instructions from the Kubernetes API Master Server.
      • Also for running pods on nodes.
      • Contains a sub-component called cAdvisor. Stands for Container Advisor.
        • This is responsible for collecting metrics from pods and exposing them with the kubelet API. That then gets sent to the Metrics Server
  • If you are using minikube, then run minikube addons enable metrics-server
  • For other environments, run the following git clone command: 
    git clone https://github.com/kubernetes-incubator/metrics-server.git
  • Deploy the required components with the kubectl create -f command, such as kubectl create -f deploy/1.8+/
    • It deploys roles, pods and services to allow the Metrics Server to pole data from the cluster.
    • Performance data can then be viewed by running kubectl top node
      • Shows CPU and Memory usage for nodes.
    • kubectl top pod shows performance metrics of pods.

Updated: