Kubernetes Scheduler

Description:

• Knows the state of all nodes
• Can define affinity parameters
  • when a group of pods should run on a same node
  • or anti-affinity

kube-scheduler:

.

• default scheduler for Kubernetes, custom scheduling component can be implemented
• select optimal Kubernetes Node to run newlly created or not yet scheduled pods
  • pods can be requirements → nodes that meet requirements are called feasible nodes
  • factors: individual and collective resource req, hardward/software/policy constraints, affinity and anti-affinity, data locality, inter-workload interference,…
• Doesnt launch the pod to node

Node selection in kube-scheduler

• select a node in 2 steps:
  1. filtering: finds the set of Nodes where it’s feasible
     • by default, it doesnt select Kubernetes Control Plane to be feasible but can be overridden
  2. scoring: ranks the remaining nodes to choose the most suitable Pod placement
• 2 supported ways to configure filtering and scoring:

---

API:

• Scheduling configuration