💡 Quick Answer: Services get DNS: <service>.<namespace>.svc.cluster.local. Same namespace: just <service>. Headless services (clusterIP: None) return individual pod IPs. Pod DNS: <pod-ip-dashed>.<namespace>.pod.cluster.local. StatefulSet pods: <pod-name>.<headless-service>.<namespace>.svc.cluster.local. Tune ndots: 2 in dnsConfig to reduce DNS lookups for external domains.

The Problem

Kubernetes creates DNS records automatically, but:

What’s the full DNS name format?
How do pods discover services in other namespaces?
How does headless service DNS differ?
Why are external DNS lookups slow (ndots)?
How to debug DNS resolution failures?

The Solution

Service DNS Records

# Service DNS format:
# <service-name>.<namespace>.svc.cluster.local

# Same namespace — short name works
curl http://my-service:8080

# Cross-namespace — need namespace
curl http://my-service.production:8080

# Full FQDN (trailing dot = absolute)
curl http://my-service.production.svc.cluster.local.:8080

# SRV records (for port discovery)
# _<port-name>._<protocol>.<service>.<namespace>.svc.cluster.local
dig SRV _http._tcp.my-service.production.svc.cluster.local
# Returns: port and target hostname

Headless Services (clusterIP: None)

apiVersion: v1
kind: Service
metadata:
  name: db-headless
  namespace: production
spec:
  clusterIP: None              # Headless!
  selector:
    app: postgres
  ports:
  - port: 5432

# Normal Service DNS → returns ClusterIP
nslookup my-service.production.svc.cluster.local
# Address: 10.96.45.123  (ClusterIP)

# Headless Service DNS → returns ALL pod IPs
nslookup db-headless.production.svc.cluster.local
# Address: 10.244.1.5    (Pod 1)
# Address: 10.244.2.8    (Pod 2)
# Address: 10.244.3.12   (Pod 3)

# Clients get all IPs and can load-balance themselves
# Essential for databases, Kafka, Elasticsearch

StatefulSet DNS

apiVersion: v1
kind: Service
metadata:
  name: postgres
spec:
  clusterIP: None
  selector:
    app: postgres
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: postgres
spec:
  serviceName: postgres        # Links to headless service
  replicas: 3

# Each StatefulSet pod gets a stable DNS name:
# <pod-name>.<service-name>.<namespace>.svc.cluster.local

nslookup postgres-0.postgres.production.svc.cluster.local
# Address: 10.244.1.5   (always postgres-0)

nslookup postgres-1.postgres.production.svc.cluster.local
# Address: 10.244.2.8   (always postgres-1)

nslookup postgres-2.postgres.production.svc.cluster.local
# Address: 10.244.3.12  (always postgres-2)

# Stable identity — even after pod restart, same DNS name
# Perfect for: primary/replica discovery, clustering

Pod DNS Records

# Pod DNS (auto-created):
# <pod-ip-dashed>.<namespace>.pod.cluster.local

# Pod 10.244.1.5 in production namespace:
nslookup 10-244-1-5.production.pod.cluster.local
# Address: 10.244.1.5

# Custom hostname and subdomain

apiVersion: v1
kind: Pod
metadata:
  name: my-pod
spec:
  hostname: my-host            # Custom hostname
  subdomain: my-subdomain      # Creates DNS record
  containers:
  - name: app
    image: nginx

# DNS: my-host.my-subdomain.<namespace>.svc.cluster.local

DNS Policy and Config

apiVersion: v1
kind: Pod
metadata:
  name: app
spec:
  dnsPolicy: ClusterFirst      # Default: use CoreDNS
  # dnsPolicy: Default         # Use node's DNS
  # dnsPolicy: None            # Only use dnsConfig
  # dnsPolicy: ClusterFirstWithHostNet  # For hostNetwork pods
  
  dnsConfig:
    nameservers:
    - 8.8.8.8                  # Additional nameserver
    searches:
    - production.svc.cluster.local
    - svc.cluster.local
    options:
    - name: ndots
      value: "2"               # Reduce DNS lookups!

The ndots Problem

# Default ndots: 5
# DNS search list: default.svc.cluster.local, svc.cluster.local, cluster.local

# Lookup "google.com" (1 dot, < 5):
# 1. google.com.default.svc.cluster.local   → NXDOMAIN
# 2. google.com.svc.cluster.local           → NXDOMAIN
# 3. google.com.cluster.local               → NXDOMAIN
# 4. google.com.                            → resolved! ✅
# = 4 DNS queries for one external lookup!

# Fix: set ndots: 2 (or use FQDN with trailing dot)
dnsConfig:
  options:
  - name: ndots
    value: "2"

# Now "google.com" (1 dot, < 2):
# 1. google.com.default.svc.cluster.local   → NXDOMAIN
# 2. google.com.                            → resolved! ✅
# = 2 queries (still not ideal)

# Best: use trailing dot for external
# curl http://api.example.com.   ← absolute, 1 query

ExternalName Services

apiVersion: v1
kind: Service
metadata:
  name: external-db
spec:
  type: ExternalName
  externalName: db.example.com
# DNS: external-db.<namespace>.svc.cluster.local → CNAME db.example.com
# No proxy — just DNS alias

Debug DNS

# Quick DNS test
kubectl run dnstest --image=busybox:1.36 --rm -it --restart=Never -- \
  nslookup my-service.production.svc.cluster.local

# Check resolv.conf
kubectl exec my-pod -- cat /etc/resolv.conf
# nameserver 10.96.0.10
# search default.svc.cluster.local svc.cluster.local cluster.local
# options ndots:5

# Detailed DNS query
kubectl run dnstest --image=nicolaka/netshoot --rm -it --restart=Never -- \
  dig my-service.production.svc.cluster.local

# Check CoreDNS
kubectl get pods -n kube-system -l k8s-app=kube-dns
kubectl logs -n kube-system -l k8s-app=kube-dns --tail=20

Common Issues

Cross-namespace service not resolving

Use full name: <service>.<namespace> or FQDN. Short names only work within the same namespace.

Slow external DNS lookups

ndots:5 causes multiple search domain lookups. Set ndots: 2 or use FQDN with trailing dot.

DNS returns stale pod IPs

CoreDNS cache. Default TTL is 30s. Check: kubectl get configmap coredns -n kube-system -o yaml.

Best Practices

Same namespace: short name — my-service:8080
Cross-namespace: include namespace — my-service.production:8080
Set ndots: 2 for pods making external calls — reduces DNS queries
Use headless services for StatefulSets and databases
Trailing dot for external — api.example.com. avoids search domain queries

Key Takeaways

Service DNS: <service>.<namespace>.svc.cluster.local
Headless services return pod IPs instead of a single ClusterIP
StatefulSet pods get stable DNS: <pod>.<service>.<namespace>.svc.cluster.local
Default ndots:5 causes slow external lookups — tune to 2
ExternalName services create CNAME records to external endpoints

K8s DNS for Services: Resolution Guide