SIG Scheduling Deep Dive
summary
Kubernetes scheduler maintains a centralized cache updated from API events and picks one pod at a time, running a feasibility filter, ranking nodes with priority functions, then binding by updating the pod spec and letting kubelet confirm. Features shown include node affinity, pod affinity/anti affinity, taints and tolerations to reserve GPUs, and priority and preemption which evicts lower priority pods to admit higher ones. Priority is alpha since 1.8, slated for beta in 1.11 and GA around 1.12; a pluggable scheduling framework and gang scheduling are in development.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:49placing pods on nodes
- 1:29caches a lot of information about the state of the cluster
- 7:13target a specific set of nodes for your pod
- 16:02raise the resource utilization of each individual node in the cluster
related talks
36mDeep Dive: SIG Scheduling
30mDeep Dive: Scheduling SIG
30mSIG Scheduling Intro & Updates
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?
29mSIG Scheduling Deep Dive
39mIntro: Scheduling SIG
25mSIG Scheduling Intro and Deep Dive
42mDeep Dive Into the Latest Kubernetes Scheduler Features