Everything You Ever Wanted to Know About Resource Scheduling, But Were Afraid to Ask

on YouTube ↗43m

" summary

Explains why CPU and memory are handled as vectors not boxes, shows how requests versus limits create QoS bands, and gives concrete failure modes like stranded resources and a DNS outage when a node dies. Describes production techniques used in Borg including Borg autopilot (used by roughly two thirds of users) and the need for a vertical pod autoscaler, cites about 800 man-years left to implement desired features, and calls for a smarter rescheduler and priority/chargeback model.

AI-written summary · about

" chapters (43)

1. 0:00Introduction
2. 1:15Warning
3. 1:32What are resources
4. 2:58What are nodes
5. 3:40Mental model of resources
6. 3:57Pack stuff into the box
7. 4:18This is really weird
8. 4:50Example
9. 5:54The Problem
10. 8:10Stranded Resources
11. 9:10The Fun Part
12. 9:23The Wrong Questions
13. 9:39Multiple Replicas
14. 9:51Saving Machines
15. 10:10Partitioning Clusters
16. 10:53Isolation
17. 12:57Memory Leaks
18. 13:30Memory Limits
19. 15:05State of Resources
20. 15:21Resource Taxonomy
21. 16:47Resource Limits
22. 17:55Quality of Service
23. 18:57Running Out of CPU
24. 20:45Coupled Resources
25. 20:55Memory
26. 22:53What happens if you dont specify
27. 23:38How much resources you need
28. 24:59No benchmarks are incredibly hard
29. 25:22Horizontal scaling
30. 26:05Calculus
31. 26:35What can we do
32. 28:37Resources cost money
33. 29:02Show of hands
34. 29:26How to do better
35. 29:44People are inherently cautious
36. 30:12We need priority
37. 32:44Overcommit model
38. 33:53Customer experience
39. 34:14Everything went sideways
40. 34:37The reality check
41. 35:53Wrapping it up
42. 36:18We have a long way to go
43. 37:19Designing applications for Kubernetes

" related talks

• 40mLimits, Requests, QoS... Rethinking What You Thought You Knew About Scheduling in KubernetesDenis Germain, Quentin JOLY · DevFest Nantes 2025
• 45mLimits, Requests, QoS, PriorityClasses: Demystifying Scheduling in K8s!Denis Germain, Quentin JOLY · Devoxx France 2026
• 21mDemystifying Resource Management in KubernetesRafael Brito, Renato Rosa · KCD Rio de Janeiro 2025
• 05mDoughnut Dilemma: A Lesson in Resource ManagersRavi Lachmann · DevOpsDays Raleigh 2019
• 24mResource Requests and Limits Under the Hood: The Journey of a Pod SpecKaslin Fields, Kohei Ota · KubeCon + CloudNativeCon Europe 2021
• 34mPatterns for building Highly Available Cluster SchedulersDiptanu Gon Choudhury · ScaleConf Colombia 2018
• 26mResource as You Wish: Collaborative Reservation and Allocation b...Takuya Mishina, Tatsuhiro Chiba · KubeDay Japan 2024
• 34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?Marek Grabowski · CloudNativeCon + KubeCon Europe 2017