Kubernetes Engine support for observation of the health and resource usage of worker nodes and key Kubernetes add-ons

You can now observe the health and resource usage of worker nodes and key Kubernetes add-ons (such as CoreDNS and KubeProxy) in clusters that you create with Kubernetes Engine. By collecting operational and resource metrics, you can now proactively detect and troubleshoot potential issues and anomalies before workloads are affected.

To enable you to observe health and resource usage, Kubernetes Engine deploys the following two managed add-ons on each worker node:

• ObservabilityAgent: The ObservabilityAgent add-on collects infrastructure metrics from sources including kubelet, cAdvisor, CoreDNS, and KubeProxy.
• NodeProblemDetector: The NodeProblemDetector add-on identifies and surfaces issues such as resource saturation and network problems at the worker node level. The managed NodeProblemDetector configuration is designed to avoid conflicts with any existing deployments of this tool, or similar tools.

These observability add-ons are pre-configured to use minimal resources, and do not require extensive custom configuration. They are assigned a low scheduling priority, ensuring that system and user workloads remain the primary focus of worker node resources. Resources for the two observability add-ons are added to the kube-system namespace.

For more information, see Observing Worker Node and Add-On Health.