Proactive cluster autoscaling in Kubernetes

on YouTube ↗29m

" summary

Shows how Kubernetes Cluster Autoscaler and Horizontal Pod Autoscaler are aligned using Pod Priority and a placeholder pod to pre-warm capacity so scale-up happens in seconds instead of minutes. Demo uses Locust traffic to prove a cold-node path of ~5 minutes versus a warmed path taking 12 seconds. Also shows a tiny sizing tool that calculates node vs pod requests to avoid premature autoscaling, and proves the placeholder approach works while accepting the cost of an underutilized pre-warmed node.

AI-written summary · about

" chapters (7)

1. 0:00Introduction to proactive autoscaling
2. 4:03How Kubernetes works
3. 8:19Scheduler and resource requests
4. 14:40Provisioning delays in autoscaling
5. 18:23Reactive vs proactive scaling
6. 22:18Placeholder pods in practice
7. 26:51Summary and trade-offs

" related talks

• 29mProactive cluster autoscaling in KubernetesChris Nesbitt-Smith · Conf42 DevSecOps 2022
• 28mCluster autoscaling in KubernetesChris Nesbitt-Smith · Conf42 Cloud Native 2023
• 34mIntroduction to AutoscalingGuy Templeton, Joe Burnett · KubeCon + CloudNativeCon North America 2020 Virtual
• 30mAutoscaling in KubernetesMarcin Wielgus · CloudNativeCon + KubeCon Europe 2017
• 35mAutoscaling Kubernetes Deployments: A (Mostly) Practical GuideNatalie Serrino · KubeCon + CloudNativeCon Europe 2022
• 34mPredictive autoscaling patterns in KubernetesRoberto Carratala, Germán M Yébenes · ContainerDays 2022
• 35mIntroduction to AutoscalingMarcin Wielgus, Vivek Bagade · KubeCon + CloudNativeCon North America 2019
• 31mFailing to Autoscale?Erwin Daria · SREday 2023 London