1.15 Kubernetes Endpoint Definition
Kubernetes Endpoint Definition explains how Kubernetes manages network access to pods through endpoints, linking services to their targets in the cluster.
Kubernetes Endpoint Definition is the precise characterization of an Endpoints or EndpointSlice object as the recorded set of network addresses currently backing a Service, defined as a derived, continuously updated artifact rather than a user-authored configuration, existing specifically to bridge a Service's abstract selector to the concrete IP addresses that kube-proxy and other consumers actually route traffic toward.
Endpoints as a Derived Object
Not Authored, but Computed
An Endpoints object is defined as the computed result of evaluating a Service's label selector against the current Pods in its namespace; unlike most objects a user creates directly, its typical contents are generated and maintained automatically by a controller rather than authored by hand.
Tied to a Specific Service
Each Endpoints or EndpointSlice object is defined in relation to exactly one Service, sharing its name and existing specifically to record that Service's current backing addresses; it has no independent meaning apart from the Service it corresponds to.
Contents of an Endpoint Record
Addresses and Ports
An individual endpoint entry is defined by an IP address, corresponding to a specific backing Pod, together with one or more ports, corresponding to the ports the Service exposes; together these form the concrete routing target that traffic destined for the Service is ultimately directed toward.
Readiness as Part of the Record
Endpoint entries are further defined by a readiness status, distinguishing addresses currently eligible to receive traffic from those temporarily excluded because their backing Pod has failed its readiness checks, ensuring that not-yet-ready or unhealthy Pods are not included among active routing targets.
EndpointSlices as a Refined Definition
Splitting Large Endpoint Sets
EndpointSlices are defined as a more scalable refinement of the original Endpoints object, splitting what would otherwise be one large object into multiple smaller slices, each holding a bounded subset of the total addresses, reducing the cost of propagating updates when a Service has very many backing Pods.
Same Underlying Purpose
Despite this structural refinement, an EndpointSlice serves the exact same defining purpose as an Endpoints object: recording the concrete, currently valid addresses that satisfy a given Service's selector.
What Endpoint Objects Are Not
Not a User-Facing Configuration Surface
Because Endpoints and EndpointSlices are generated automatically for selector-based Services, they are not intended as a primary configuration surface; manual endpoint management is defined as relevant only for the narrower case of selector-less Services pointing at external or manually managed addresses.