Nodeless Knative: Just-in-Time All the Way

on YouTube ↗22m

" summary

Shows a working nodeless stack that provisions real kubelet worker nodes on demand (demo uses Luna on EKS and kind clusters), schedules pods only after a node spins up, and tears nodes down when idle. Uses Kubernetes primitives plus vertical pod autoscaler to right‑size requests and falls back to smallest instance then upsizes on later runs. Demonstrates multi‑cluster federation that brings up control planes on demand (clusters enter standby after ~2 minutes). Cold starts occur on first call; subsequent requests succeed.

AI-written summary · about

" chapters (8)

1. 0:00Nodeless Knative intro
2. 1:11Nodeless concept single cluster
3. 3:39Live demo JIT compute
4. 7:15Q&A single-cluster setup
5. 11:54Extending to multi-cluster
6. 13:42Live demo multi-cluster JIT
7. 16:37Summary takeaways
8. 18:42Q&A multi-cluster scheduling

" related talks

• 39mDelivering Serverless Experience on Kubernetes: Beyond Web ApplicationsAlex Glikson · KubeCon + CloudNativeCon Europe 2019
• 31mBeyond Traditional DevOps: Dive into Serverless with K8sKaran Jagtiani · Conf42 Kube Native 2023
• 55mServerless On Your Own Terms using KnativeMark Chmarny · JAX London 2019
• 39mServerless Platform for Large Scale Mini-Apps: From Knative to ProductionYitao Dong, Ke Wang · KubeCon + CloudNativeCon North America 2019
• 52mServerless with KnativeMete Atamel · NDC London 2019
• 20mKnative: A Kubernetes Framework to manage Serverless WorkloadsNikhil Barthwal · DevOps Pro Europe 2019
• 23mServerless with KnativeMete Atamel · FOSSASIA Summit 2019
• 31mKnative: Is it for me?Elad Hirsch, Kornel Chlebovics · KCD Israel 2023