Architect Using KEDA, Helm, Virtual-Kubelet to Save Cost

on YouTube ↗39m

" summary

Claim: using KEDA with HPA plus Virtual Kubelet and Helm cuts a big ML batch job from ~25 lakh to ~3.5–4 lakh (about 1 lakh was log cost). Setup uses KEDA + HPA to scale by queue depth (RabbitMQ/Kafka scalers), deploys an Azure Function container via a Helm chart scaledObject, and runs pods on Virtual Kubelet on ACI to pay per 0.1 CPU / 0.1 GB RAM. Demo shows queued messages trigger KEDA, HPA creates pods on virtual node, and work completes without full node billing.

AI-written summary · about

" chapters (11)

1. 0:00Introduction
2. 2:56Agenda and autoscaling
3. 5:27Kubernetes HPA limits
4. 7:12KEDA scaling solutions
5. 9:21Virtual Kubelet cost savings
6. 13:32Virtual Kubelet deep dive
7. 19:30Helm for Kubernetes
8. 20:16KEDA and Virtual Kubelet demo
9. 30:26Summary
10. 31:31Book promotion
11. 34:29Q&A

" related talks

• 48mArchitect using KEDA/Helm/Virtual-Kubelets to save costKarthikeyan V K · AzConf 2020
• 43mKubernetes cost optimisation - How to actually do itAnnie Talvasto · Build Stuff 2021
• 31mScaling Your Cluster (Both Ways)Patrick Chanezon, Scott Coulton · KubeCon + CloudNativeCon North America 2019
• 33mKubernetes cost optimization - How to actually do itAnnie Talvasto · Automation + DevOps Summit 2021
• 38mScaling EDA Workloads with Kubernetes, KEDA & KarpenterNatasha Wright · GOTO EDA Day London 2024
• 46mScaling EDA Workloads with Kubernetes, KEDA & KarpenterNatasha Wright · GOTO EDA Day London 2024
• 37mKubernetes Event-driven Autoscaling with KEDAZbynek Roubalik, Jorge Turrado · KubeCon + CloudNativeCon Europe 2022
• 35mKEDA - Real Time And Serverless Scaling In KubernetesZbynek Roubalik, Jeff Hollan · KubeCon + CloudNativeCon North America 2022