Building a Kubernetes Scheduler using Custom Metrics
summary
Shows how to replace scheduling decisions with a custom scheduler that queries monitoring backends (uses statsd, JMX, Prometheus and a Sysdig demo) and posts bind operations to the API; includes runnable source and placeholder containers. Implementation runs outside the cluster for testing but is meant to run as a pod, implements caching and fallback to delegate to the default scheduler on error, demonstrates parent-object edits (deployment) to force reschedule, and discusses preemption, affinities and failure modes.
AI-written summary · about
chapters (21)
- 0:00Intro
- 0:18Building a Kubernetes Schedules using Custom Metrics
- 1:38This talk is not about
- 2:12Schedulers, Scalers ...
- 4:03Birth of a pod
- 5:10Zooming in - The scheduler job
- 5:47Scheduler - Basic Behaviour
- 7:12Scheduling - Hard constraints
- 7:41Scheduling - Soft constraints
- 9:04Scheduling - Affinity Complexity
- 9:36Scheduling: Experimental features
- 11:34Why custom metrics
- 14:20Pod scheduler - RBAC credentials
- 14:53Two different Binding resources in the API
- 15:38Building a Kubernetes Scheduler using Custom Metrics
- 19:00Creating your own scheduler... WHAT COULD POSSIBLY
- 19:28Improvements - Metrics cache
- 20:15Improvements - Failsafe plans
- 24:22Improvements - Honor constraints
- 25:08Improvements - Modify Pod YAML
- 25:51Pod scheduler - Improvements V
related talks
01h 29mTutorial: Unleash the Full Potential Of Kubernetes Scheduler: Configuration, Extension And Operation In Production
25mKube-scheduler-evaluator: Evaluating Kubernetes Schedulers at Hyperscaler…
34mWorking your Cluster: Smarter Scheduling Decisions for Your Work...
36mA Toolkit for Simulating Kubernetes Scheduling at Scale
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?
27mUnderstanding Kube-Scheduler-Simulator
36mDeep Dive: SIG Scheduling
01h 01mKubernetes – going beyond the basics