GKE Autopilot Presentation Notes

Introduction

• Speaker: William Dennis, Product Manager on Google Cloud
• Topic: GKE Autopilot mode for application delivery and management
• Co-presenter: Gary (will present demos)

Deployment Options

• Traditional VM deployment using GCE
• Serverless deployment with Cloud Run
• Kubernetes with GKE lies in between
• Trade-offs:
  • Higher abstraction = easier management
  • Less flexibility in deployment choices

Why Kubernetes?

• Popular for handling complex requirements:
  • Legacy applications
  • Complex deployments needing persistent disks
• Power of Kubernetes:
  • Flexible and practical deployment environment
  • Open-source technology for portability
  • Runs on-premise or in different clouds

Kubernetes Components

• Out-of-the-box components include:
  • Deployment for stateless web apps
  • Stateful sets for custom databases
  • Job objects for batch jobs
  • Daemon sets for agents on nodes
• Scheduling constructs:
  • Zonal affinity
  • Pod topology spread patterns
  • Priority and preemption for scaling high-priority workloads

Learning Curve

• Steep learning curve is acknowledged, but:
  • Simple deployments require only learning two constructs: Deployment and Service
  • Powerful tools are necessary for complex deployments
• Challenges:
  • Need to understand both Kubernetes architecture and GKE platform APIs

Introduction of Autopilot

• Autopilot versus Standard mode of GKE:
  • Autopilot simplifies cluster creation and management
  • Automates many platform details
• Cluster creation process is simplified:
  • Minimal configuration needed (cluster name, region, network settings)
  • No node or auto-scaling setup required

Benefits of Autopilot

• Provisioning of node resources is automated based on workload requirements.
• Pod-level SLA with three nines availability.
• Focus on running applications rather than managing infrastructure.
• Billing model based on pod requests, not node resources:
  • No need for Kubernetes bin packing expertise
  • Easier cost tracking for multi-team setups

Security and Management

• Strong security posture with GKE hardening guidelines
• Automatic updates and maintenance options available
• Pod-level constructs (e.g., Deployments, StatefulSets) still function normally in Autopilot

User Case: Ubi

• Company: Ubi, a medical technology startup in Japan
• Benefits seen with Autopilot:
  • Focus on healthcare solutions rather than cluster management

Demo by Gary

• Objective: Deploy a standard web application with a Redis backend on Autopilot
• Real-time provisioning of compute resources by Autopilot
• Successful deployment confirmed with all pods running

Scaling Workloads

• Scaling options available in GKE Autopilot:
  • Vertical Pod Autoscaler (VPA): adjusts pod sizes based on resource utilization
  • Horizontal Pod Autoscaler (HPA): adjusts the number of pod replicas based on demand
• HPA can use various metrics, including CPU and custom metrics

Second Demo by Gary

• Objective: Autoscale a PubSub workload using HPA based on metrics from Google Cloud Monitor
• Steps:
  • Setup of workload identity for authentication
  • Created a namespace, service account, and PubSub service account
  • Deployment of application and HPA resource configured for unacknowledged messages
• Demonstrated HPA adjustment in response to load

Conclusion

• GKE Autopilot significantly reduces overhead in managing Kubernetes clusters
• Encouragement to explore further through provided links.