How to configure metrics-server to monitor kubernetes cluster resources ?

For any virtualization or cloud platform, resource monitoring is a critical operational component. And its applicable for Kubernetes platform as well. As documented in Tools for Monitoring Resources, a lightweight, short-term, in-memory tool called metrics-server cab be used to collect the CPU & Memory utilization of nodes and pods in a kubernetes cluster.

The configuration of metrics-server is really simple and can be deployed easily on the kubernetes-master node. Refer the following steps to configure the metrics-server on an existing kubernetes cluster:

a. We have a 3 node cluster already deployed and ready for the configuration.

   
     

b. Login to the k8s-master node and get the components.yaml file from the latest release. While writing this document, the latest release is v0.4.2

    # wget  https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml   

c. Open the file and go to the containers argument section:

 

d. Include the following line as marked below and save the file:

     

e. Now the deployment file is ready for the deployment. Use the below command to deploy it:

  # kubectl apply -f  components.yaml

f. If anything will be available, it wont change and new components will be created:

     

g. Once deployment is complete, you can see the deployment, service and pods as mentioned below:

 

h. If you will check for the apiservices, you should be able to see a new entry there too which is running in the kube-system namespace:

     

Now it will take just few seconds to collect the nodes and pods resource utilization details. And you should be able to query the resource details using the following commands:

a. To get the node's resource utilization, use the below command:

     

b. To get the pod's resource utilization, use the below command:

     

    You can filter the namespace based resource utilization using specific flags as mentioned above.

c. To check how many containers are running in a pod, use the below command:

     

    Now as highlighted above, there is a pod in the kube-system namespace that's having two containers in it. If you want to check the resource utilization of both the containers in that pod, use the below command as shown:

     

    The command line tool is having some limitations and monitoring the resources beyond this out of scope at this point of time. That's why a better way to monitor the container level resource utilization is using Prometheus. Its a community driven opensource monitoring tool with much better monitoring capabilities. Refer the architecture documents to know more about it.