Autopilot Became the Default Operation Mode for Google Kubernetes Engine

Published on may 2, 2023

For GKE clusters, Google has announced that Autopilot is now the default and recommended operation mode.

Autopilot was introduced in 2021 as a way to manage Kubernetes clusters. Autopilot performs all management tasks based on the best practices learned from Google SRE and engineering.

GKE consists of a control plane and worker nodes. The following diagram illustrates how Autopilot manages the entire infrastructure. In GKE standard mode, Google manages the control plane and the system components, and the users manage the nodes.

Developers are abstracted from the GKE cluster management by Autopilot, and the cluster infrastructure is provisioned based only on the workload. To model consumption, Autopilot uses the concept of compute classes, which are a subset of the GCP Compute Engine machine series. As a default, GKE Autopilot PODs run on a computing platform optimized for general-purpose workloads, such as web serving and medium-intensity batch processing. As a result of the concept of compute classes, developers are able to specify specific resources, and CPU platforms in a workload definition. It automatically spins up the infrastructure and sets the appropriate taints and tolerances based on the workload.

Using Autopilot, the control plane of the cluster is constantly monitored by Google to ensure that PODs are always scheduled and scaled appropriately. Google is also concerned about security issues. With Autopilot, the security-focused version of Kubernetes is used and security best practices are applied. By default, the shielded node is used.

Autopilot allows users to pay only for the effective POD resources requested in PodSpecs, and no additional infrastructure costs are required, as opposed to a traditional Kubernetes cluster where some resources are overprovisioned (for different reasons) and cost even if they are not used. Kubernetes allocates some of the necessary resources for system workloads on each node, and these resources are billed to customers. Because Autopilot only charges for the resources specified in the podSpecs, this problem is eliminated.