Skip to main content
Kubernetes Services provide a stable network endpoint for Pods, but under the hood, Services rely on two resources—Endpoints and Endpoint Slices—to map Service selectors to Pod IPs and ports. This guide explains how each resource works, their roles in traffic routing, and how to inspect them.

​
Endpoints

An Endpoint object in Kubernetes keeps track of all Pod IPs and ports that back a Service. When you create a Service, Kubernetes uses its label selector to identify matching Pods and generates an Endpoints object with the same name.
The image illustrates a network diagram showing a service (SVC) connected to endpoints with specific IP addresses, which in turn connect to three different pods.
As Pods are created, deleted, or updated, the Endpoints object is managed automatically by the control plane—no manual edits are needed.

​
Inspecting an Endpoints Object

Use kubectl to view the Endpoints backing a Service:
Example output:
Fields explained:
  • subsets: Groups of addresses and ports.
  • addresses: Lists Pod IPs and node names.
  • ports: Indicates which port names and protocols are exposed.

​
Limitations of Endpoints

Each Endpoints object can hold up to 1,000 addresses. In large clusters or high-scale environments, this cap may lead to performance bottlenecks or truncated endpoint lists.
ResourceMaximum AddressesImpact
Endpoints1,000May require Service sharding or slicing
The image is a slide titled "Endpoint – Limitations," indicating that Kubernetes restricts endpoints to 1,000.

​
Endpoint Slices

Introduced as a beta feature in v1.17 and GA in v1.21, Endpoint Slices solve the scalability issues of Endpoints by splitting Pod endpoints into multiple, smaller objects. Each slice holds a subset of IPs, ports, and metadata, reducing control-plane load.
The image is a diagram titled "Endpoint Slices," illustrating an advanced, scalable way to manage a network, with a network icon and a gradient label.
Endpoint Slices support optional fields like topology hints, enabling you to optimize traffic routing based on node labels or zones.

​
How Endpoint Slices Work

The image illustrates how Endpoint Slices work in Kubernetes, showing the relationship between services, endpoint slices, and pods, with notes on management and dynamic updates.
  • Slices per Service: Multiple Endpoint Slice objects can coexist for one Service.
  • Subset Size: By default, each slice holds up to 100 endpoints, configurable via the EndpointSlice API.
  • Automatic Updates: The control plane synchronizes slices as Pods change state.

​
Viewing Endpoint Slices

Sample YAML:
FeatureEndpointsEndpoint Slices
Scalability≤1,000 IPsConfigurable slice size, multiple slices per Service
Metadata & TopologyLimitedSupports zone hints, custom labels
Control-Plane LoadHigherDistributed across slices, fewer large updates overall

​
References

Watch Video