How Scheduling in Kubernetes Works
summary
Scheduler watches pod creations, filters then scores candidate nodes (default 40% of nodes are scored), then creates a binding the kubelet acts on. Demo on Kubernetes v1.16 with nodes named Lannister, Stark and Targaryen shows hard constraints via nodeSelector/nodeAffinity, soft preferences via preferredDuringScheduling podAffinity, and taints/tolerations (NoSchedule/NoExecute) causing pods to pend or terminate until tolerated. Also notes the scheduling framework (alpha v1.15), pod spreading and a move toward batch scheduling to place multiple pods at once.
AI-written summary · about
chapters (3)
related talks
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?
35mSIG Scheduling Deep Dive
38mHow Kubernetes Plays Tetris with Your Containers
42mDeep Dive Into the Latest Kubernetes Scheduler Features
30mDeep Dive: Scheduling SIG
36mDeep Dive: SIG Scheduling
24mK8s Pod Scheduling - Deep Dive
58mHow Kubernetes plays Tetris with your containers