An EndpointSlice is a Kubernetes API resource that represents a scalable set of network endpoints for a Service. It divides one Service’s endpoints across multiple objects, avoiding the performance cost of sending one large Endpoints object to every cluster node whenever a change occurs. Kubernetes creates and manages these slices automatically. section 11.3.2
EndpointSlices make Service endpoints easier to scale and inspect:
- Each slice contains one endpoint subset; different Service ports can therefore appear in different slices. section 11.3.2
- A slice supports up to 1,000 endpoints, while Kubernetes adds up to 100 endpoints per slice by default; ports are limited to 100. section 11.3.2
- Slice names include a generated suffix, so one Service can have several independently named EndpointSlices. section 11.3.2
- Endpoint data records addresses, readiness, the target Pod, node, and available ports, making it useful for examining why traffic does or does not reach a Pod. section 11.3.2 section 11.6.2
List and inspect a Service's slices
Run `kubectl get endpointslices` to list slices. Filter them with `kubectl get endpointslices -l kubernetes.io/service-name=kiada`, then inspect one with `kubectl describe endpointslice kiada`. The output shows IPv4 addressing, ports such as 8080 and 8443, endpoint addresses, a `ready` condition, and the referenced Pod.
Readiness and topology-aware routing
A Pod with an unsuccessful readiness probe remains represented in its EndpointSlice with `conditions.ready: false`, so the Service should not send it traffic. When topology-aware hints are enabled with the Service annotation `service.kubernetes.io/topology-aware-hints: Auto`, the EndpointSlice can identify zones that should consume each endpoint, helping keep traffic within an availability zone when suitable endpoints exist. section 11.6.2 section 11.5.2
