HPA is Not the Answer - Why Response Time SLOs should drive the Kubernetes platform and underlying Infrastructure

Customer Challenges

• Kubernetes provides Horizontal Pod Autoscaler (HPA) for autoscaling container-based stateless applications
  
  
• HPA has limitations
  • It requires lot of work from Developers to determine whether CPU or Memory is the best resource on which to define the policies and a lot of work to set the CPU/Memory thresholds that will trigger the scaling. And THIS HAS TO BE DONE FOR EVERY SERVICE
  • CPU and Memory are utilisation metrics - Not direct indicators of app performance