Deployments and YAML, Microsoft AZ-304 Certifications, Public IP addresses, What to watch out for

Nodes and node pools – Designing Compute Solutions

by Katie Campbell

An AKS cluster has one or more nodes, which are virtual machines running the Kubernetes node components and container runtime:

  • kubelet is the Kubernetes agent that responds to requests from the cluster master and runs the requested containers.
  • kube-proxy manages virtual networking.
  • The container runtime is the Docker engine that runs your containers.

The following diagram shows these components and their relation to Azure:

Figure 7.12 – AKS nodes

When you define your AKS nodes, you choose the SKU of the VM you want, which in turn determines the number of CPUs, RAM, and type of disk. You can also run GPU-powered VMs, which are great for mathematical and AI-related workloads.

You can also set up the maximum and the minimum number of nodes to run in your cluster, and AKS will automatically add and remove nodes within those limits.

AKS nodes are built with either Ubuntu Linux or Windows 2019, and because the cluster is managed, you cannot change this. If you need to specify your OS or use a different container runtime, you must build your Kubernetes cluster using the appropriate engine.

When you define your node sizes, you need to be aware that Azure automatically reserves an amount of CPU and RAM to ensure each node performs as expected – these reservations are 60 ms for CPU and 20% of RAM, up to 4 GB So, if your VMs have 7 GB RAM, the reservation will be 1.4 GB but for any VM with 20 GB RAM and above, the reservation will be 4 GB.

This means that the actual RAM and CPU amounts available to your nodes will always be slightly less than the size would otherwise indicate.

When you have more than one node of the same configuration, you group them into a node pool, and the first node is created within the default node pool. When you upgrade or scale an AKS cluster, the action will be performed against either the default node pool or a specific node pool of your choosing.

Pods

A node runs your applications within pods. Typically, a pod has a one-to-one mapping to a container, that is, a running instance. However, in advanced scenarios, you can run multiple containers within a single pod.

At the pod level, you define the number of resources to assign to your particular services, such as the amount of RAM and CPU. When pods are required to run Kubernetes, the scheduler attempts to run the pod on a node with available resources to match what you have defined.