Say hi to the new GKE: A better default compute for every workload

Key Concepts:

• GKE (Google Kubernetes Engine) Standard Mode
• Autopilot Mode
• Compute Classes (Autopilot, Autopilot Spot, Custom)
• Node Selectors
• Autopilot Managed Nodes
• User Managed Nodes
• GCloud CLI
• Kube Control CLI
• Pod-based pricing

1. Introduction of GKE Standard Mode with Autopilot Workloads

• The video introduces the new capability of running Autopilot workloads within GKE Standard Mode clusters.
• Previously, Autopilot and Standard were mutually exclusive cluster creation choices.
• Now, GKE Standard offers the flexibility to run both standard and Autopilot workloads within the same cluster.

2. Creating a GKE Standard Cluster Compatible with Autopilot

• Demonstration: The video demonstrates creating a GKE Standard cluster in the us-east1 region with three nodes.
• Console Reminder: The Google Cloud Console provides a reminder that the standard cluster is compatible with Autopilot compute classes.
• Compatibility Features: Specific features need to be enabled for a standard cluster to be compatible with Autopilot compute classes. These settings are enabled by default.
• Release Channel: Changing the target release channel (e.g., from Rapid or Regular to Stable) can make the cluster incompatible with Autopilot compute classes, triggering a warning in the console. The video switches the release channel back to "regular" to maintain compatibility.
• Command-Line Creation: The cluster can also be created using the gcloud CLI or Infrastructure as Code (IaC) tools.

3. Verifying Cluster Readiness and Compute Classes

• Cluster Status: After creation, the cluster is shown as ready in the Cloud Console.
• kubectl Access: The kubectl CLI is connected to the cluster.
• Node Verification: The kubectl get nodes command confirms the presence of the three nodes.
• Compute Class Verification: The kubectl get computeclass command shows that the autopilot and autopilot-spot compute classes are already enabled. These compute classes are used to instruct GKE to run workloads in Autopilot mode.

4. Deploying Workloads with Node Selectors

• Workload Definition: Two almost identical workloads are presented. The key difference is the nodeSelector in one workload.
• nodeSelector: The workload intended for Autopilot has the nodeSelector set to cloud.google.com/gke-compute-class: "autopilot". This tells GKE to run the workload on an Autopilot-managed node.
• Workload Creation: Both workloads are created.

5. Observing Workload Execution and Node Types

• Workload Status: The Cloud Console shows both workloads running.
• Node Type Differentiation: The Autopilot workload runs on an "Autopilot managed" node, while the other workload runs on a "user managed" node.
• New Autopilot Node: The kubectl get nodes command reveals that a new node named gke-autopilot has been created by GKE to run the Autopilot workload.
• Pod Placement Confirmation: The kubectl get pods -o wide command confirms that the "Hello World Autopilot" pod is running on the gke-autopilot node.

6. Creating Custom Compute Classes for Autopilot

• Custom Compute Class Definition: To create a custom compute class for Autopilot, the autopilot.gke.io/enabled: "true" label must be added to the compute class specification.
• Functionality: The rest of the compute class functionality can remain the same.

7. Benefits of Autopilot Mode in GKE

• Fully Managed Experience: Autopilot mode provides a fully managed GKE experience.
• Pod-Based Pricing: Workloads in Autopilot mode are billed based on pod resource consumption.

8. Conclusion

• The video encourages users to try running Autopilot workloads in GKE Standard and provide feedback.
• The new capability offers flexibility and cost optimization for GKE deployments.

Key Quotes:

• "Autopilot Mode is optimized for running most production workloads in an environment that applies recommended settings for security, reliability, performance, and scalability."
• "With autopilot mode in gk, you can get fully managed GK experience with pod based pricing for your workload."