Discussion: CPU-based Autoscaling Strategy for KubeOne Cluster with OOM Issues

[Aj7ay7]

Current Situation

• Nodes are getting OOM (Out of Memory) killed
• Using addons and parameters in the cluster

Memory vs CPU Autoscaling

• Is it feasible to switch from memory-based to CPU-based autoscaling in KubeOne?
• What are the implications of this change?

Node Transition Handling

• How can we better manage the transition between old and new nodes?
• What strategies can we implement to maintain whitelist IP configurations during node replacement?

Resource Management

• What's the best approach to handle addons and parameters during autoscaling?
• How can we prevent OOM situations while maintaining optimal resource utilization?