Autoscaling and Progressive Delivery - A Match Made in Heaven
summary
Demonstrates that progressive delivery can run with autoscalers: authors run 10 concrete examples, including an HPA watching memory (min 1, max 10) against a test app that allocates 1MB per request, and a Linkerd traffic-management case that sends 80% to one pod. Shows two cost levers that work: previewReplicaCount (example: previewReplicaCount=3 forces three preview pods) and dynamicStableScale to shrink stable pods as canary grows. Conclusion: Argo Rollouts accepts an HPA but overrides it for pods explicitly managed by rollout logic.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:46Progressive delivery with Argo Rollouts
- 5:59Autoscaling basics
- 7:28Cost challenges
- 8:58Practical examples
- 10:02Cost control without HPA
- 12:58Traffic management for costs
- 14:05HPA with Argo Rollouts
- 16:12HPA vs Argo Rollouts conflicts
- 17:18Key takeaways
- 19:02Inclusion resources
related talks
35mProgressive Delivery with Argo Rollouts
51mProgressive Delivery in the Kubernetes era
30mProgressive Delivery Made Easy With Argo Rollouts
35mAutoscaling Kubernetes Deployments: A (Mostly) Practical Guide
26mGlow up Your Stateful Services with Progressive Delivery Using Argo Rollouts!
40mProgressive Delivery with Kubernetes and Argo Rollouts
28mCross-Cluster Progressive Delivery: Rolling Out Apps With Arg...
46mProgressive Delivery in the Kubernetes era