1.24 Kubernetes Scheduling Definition

Kubernetes Scheduling Definition is the precise characterization of scheduling as the specific act of binding an otherwise unassigned Pod to exactly one node, defined narrowly as this single binding decision rather than the broader set of concerns, such as autoscaling or workload management, that are often discussed alongside it but fall outside the scheduler's own defined responsibility.

Scheduling Defined as a Binding Decision

One Pod, One Node

Scheduling is defined at its core as the act of writing a binding between a specific Pod and a specific node into the cluster's state; before this binding exists, a Pod is defined as unscheduled, and after it exists, the kubelet on the bound node becomes responsible for actually running the Pod's containers.

A Decision, Not an Execution

The scheduler's defined role ends at making and recording this decision; it does not itself start any containers or communicate with the container runtime, leaving execution entirely to the kubelet running on the node it selected.

Filtering and Scoring as Defined Subphases

Filtering Narrows the Candidate Set

Filtering is defined as the subphase that eliminates every node incapable of satisfying a Pod's hard requirements, such as insufficient allocatable resources or unmatched taints, producing a reduced set of feasible nodes as its output.

Scoring Selects Among Feasible Nodes

Scoring is defined as the subphase that ranks the nodes remaining after filtering according to configured preferences, such as resource balance or affinity rules, ultimately selecting the single highest-scoring node as the target of the binding decision.

Scheduling as Distinct from Related Concepts

Not Autoscaling

Scheduling is defined independently of autoscaling: the scheduler places Pods onto nodes that already exist, while decisions about whether to add or remove nodes or replicas altogether belong to separate mechanisms, namely the Cluster Autoscaler and the Horizontal Pod Autoscaler.

Not Workload Management

Scheduling is likewise defined independently of workload controllers such as Deployments; a controller decides how many Pods should exist and what they should look like, while the scheduler's defined concern begins only once such a Pod already exists and needs a node assignment.

Custom and Multiple Schedulers Within the Definition

Scheduler as a Pluggable Role

The definition of scheduling permits multiple scheduler implementations to operate within the same cluster simultaneously, with each Pod specifying which scheduler is responsible for it, meaning "the scheduler" is defined as a role that can be fulfilled by different implementations rather than a single fixed component.

Scheduling Decision Diagram