EndpointSlice Service Discovery

💡 Quick Answer: EndpointSlices replace legacy Endpoints for services with many backends. They’re sharded into 100-endpoint chunks, reducing API server load and enabling topology-aware routing.

The Problem

Legacy Endpoints objects store all pod IPs in a single resource. For services with hundreds of backends, every pod change triggers a full Endpoints update — overwhelming the API server and kube-proxy with large watch payloads.

The Solution

View EndpointSlices

# List EndpointSlices for a service
kubectl get endpointslices -l kubernetes.io/service-name=web-app

# Detailed view
kubectl describe endpointslice web-app-abc12

EndpointSlice Structure

apiVersion: discovery.k8s.io/v1
kind: EndpointSlice
metadata:
  name: web-app-abc12
  labels:
    kubernetes.io/service-name: web-app
addressType: IPv4
ports:
  - name: http
    port: 8080
    protocol: TCP
endpoints:
  - addresses:
      - "10.244.1.5"
    conditions:
      ready: true
      serving: true
      terminating: false
    nodeName: worker-01
    zone: us-east-1a
  - addresses:
      - "10.244.2.8"
    conditions:
      ready: true
    nodeName: worker-02
    zone: us-east-1b

Enable Topology-Aware Routing

apiVersion: v1
kind: Service
metadata:
  name: web-app
  annotations:
    service.kubernetes.io/topology-mode: Auto
spec:
  selector:
    app: web-app
  ports:
    - port: 80
      targetPort: 8080

Custom EndpointSlice (External Service)

apiVersion: discovery.k8s.io/v1
kind: EndpointSlice
metadata:
  name: external-db
  labels:
    kubernetes.io/service-name: external-db
addressType: IPv4
ports:
  - name: mysql
    port: 3306
    protocol: TCP
endpoints:
  - addresses:
      - "192.168.1.100"
    conditions:
      ready: true
---
apiVersion: v1
kind: Service
metadata:
  name: external-db
spec:
  ports:
    - port: 3306

graph TD
    S[Service: web-app] --> ES1[EndpointSlice 1<br/>100 endpoints]
    S --> ES2[EndpointSlice 2<br/>100 endpoints]
    S --> ES3[EndpointSlice 3<br/>50 endpoints]
    ES1 --> P1[Pod IPs + zone + ready status]
    ES2 --> P2[Pod IPs + zone + ready status]
    ES3 --> P3[Pod IPs + zone + ready status]

Common Issues

EndpointSlice shows pods as not ready Check pod readiness probes:

kubectl get endpointslices -l kubernetes.io/service-name=web-app -o yaml | grep -A3 conditions

Topology-aware routing causing uneven distribution If one zone has fewer pods, traffic may overflow to other zones. Ensure sufficient replicas per zone.

Stale endpoints during rolling updates The terminating condition prevents traffic to shutting-down pods. Ensure terminationGracePeriodSeconds allows in-flight requests to complete.

Best Practices

• EndpointSlices are created automatically — no manual management needed for standard services
• Use topology-aware routing to reduce cross-zone network costs
• Monitor EndpointSlice count for very large services (>1000 pods)
• Use serving condition to distinguish ready from terminating pods
• Set publishNotReadyAddresses: true on headless services for StatefulSets that need pre-ready DNS

Key Takeaways

• EndpointSlices shard endpoints into 100-pod chunks (configurable)
• They replace legacy Endpoints for all new services (GA since 1.21)
• Each slice includes topology info (zone, node) for routing decisions
• topology-mode: Auto enables zone-preferring traffic routing
• The terminating condition enables graceful connection draining
• Manual EndpointSlices enable Kubernetes service discovery for external resources