Day 7: Services and CoreDNS
Goal: You expose apps through all Service types – including a real LoadBalancer on VMs – and you know how name resolution works inside the cluster. · Duration: approx. 60 min · Prerequisite: Day 4
Briefing
CKA: Services and Networking (20 %) – “Use ClusterIP, NodePort, LoadBalancer service types and endpoints”, “Understand and use CoreDNS”.
outside --> LoadBalancer IP (MetalLB) --+
outside --> NodeIP:NodePort ------------+--> ClusterIP --> EndpointSlice --> Pods
Pod --> echo.tag07 (CoreDNS) -------+
Each type builds on the previous one: a LoadBalancer also has a NodePort, and a NodePort also has a ClusterIP.
mkdir -p ~/cka/tag07 && cd ~/cka/tag07
k create ns tag07
k config set-context --current --namespace=tag07Drill 1: An App That Reveals Its Name
agnhost is the Kubernetes developers’ test tool;
/hostname returns the Pod name.
k create deployment echo --image=registry.k8s.io/e2e-test-images/agnhost:2.53 \
--replicas=2 -- /agnhost netexec --http-port=8080
k rollout status deployment echo
k expose deployment echo --port=80 --target-port=8080
k get svc echoExpected:
service/echo exposed
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
echo ClusterIP 10.97.170.32 <none> 80/TCP 1s
--port is the port of the Service,
--target-port the port in the container.
k run c --image=busybox:1.37 --rm -it --restart=Never -- \
sh -c 'wget -qO- echo/hostname; echo; wget -qO- echo/hostname; echo'Expected (random order):
echo-57489dc7f5-fb5kn
echo-57489dc7f5-6g2rl
Save your progress? Sign up for free
Drill 2: EndpointSlices
Behind every Service with a selector, Kubernetes maintains the list of Pod IPs:
k get endpointslices -l kubernetes.io/service-name=echoExpected:
NAME ADDRESSTYPE PORTS ENDPOINTS AGE
echo-dgrlz IPv4 8080 192.168.189.119,192.168.235.160 0s
Scale – the list grows along with it:
k scale deployment echo --replicas=3
J='{range .items[*].endpoints[*]}{.addresses[0]} {.nodeName}{"\n"}{end}'
k get endpointslices -l kubernetes.io/service-name=echo -o jsonpath="$J"Expected:
192.168.189.119 worker2
192.168.235.160 worker1
192.168.189.66 worker2
A Service without endpoints is the most common network failure – usually the selector does not match the Pod labels (Day 15).
Save your progress? Sign up for free
Drill 3: NodePort
k expose deployment echo --name=echo-np --type=NodePort --port=80 --target-port=8080
k get svc echo-npExpected:
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
echo-np NodePort 10.105.132.187 <none> 80:31482/TCP 0s
The NodePort (here 31482, from 30000–32767) is open on
every node, including cp1:
NP=$(k get svc echo-np -o jsonpath='{.spec.ports[0].nodePort}')
curl -s 172.16.0.228:$NP/hostname; echo
curl -s 172.16.0.126:$NP/hostname; echoExpected:
echo-57489dc7f5-6g2rl
echo-57489dc7f5-fb5kn
If nothing comes back right after creating it, wait a few seconds – kube-proxy needs a moment for the rules.
Save your progress? Sign up for free
Drill 4: MetalLB – LoadBalancer Without a Cloud
In the cloud, the provider supplies the external IP. On VMs, MetalLB takes over: it assigns IPs from a pool and answers ARP requests for them.
ML=https://raw.githubusercontent.com/metallb/metallb/v0.16.2/config/manifests
k apply -f $ML/metallb-native.yaml
k wait -n metallb-system --for=condition=Ready pod --all --timeout=180sExpected (abridged):
pod/controller-6b66bcf7d6-xmkhv condition met
pod/speaker-5d6vl condition met
pod/speaker-6ndkl condition met
pod/speaker-xqqpf condition met
Pick the pool from free addresses in your VM network – here 172.16.0.242–244:
cat <<'EOF' | k apply -f -
apiVersion: metallb.io/v1beta1
kind: IPAddressPool
metadata:
name: lab
namespace: metallb-system
spec:
addresses:
- 172.16.0.242-172.16.0.244
---
apiVersion: metallb.io/v1beta1
kind: L2Advertisement
metadata:
name: lab
namespace: metallb-system
spec:
ipAddressPools:
- lab
EOFExpected:
ipaddresspool.metallb.io/lab created
l2advertisement.metallb.io/lab created
Warning: The addresses must neither be in use nor lie within your router’s DHCP range. Check with
pingfirst.
Save your progress? Sign up for free
Drill 5: Service of Type LoadBalancer
k expose deployment echo --name=echo-lb --type=LoadBalancer --port=80 --target-port=8080
k get svc echo-lbExpected:
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
echo-lb LoadBalancer 10.101.11.148 172.16.0.242 80:31278/TCP 5s
From your laptop, outside the cluster:
curl -s 172.16.0.242/hostname; echoExpected: a Pod name such as
echo-57489dc7f5-fb5kn. Without MetalLB,
EXTERNAL-IP would stay <pending> forever
– a typical exam scenario.
Save your progress? Sign up for free
Drill 6: How a Pod Resolves Names
k get svc kube-dns -n kube-system
k run d --image=busybox:1.37 --rm -it --restart=Never -- cat /etc/resolv.confExpected:
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kube-dns ClusterIP 10.96.0.10 <none> 53/UDP,53/TCP,9153/TCP 47m
search tag07.svc.cluster.local svc.cluster.local cluster.local localdomain
nameserver 10.96.0.10
options ndots:5
For historical reasons, the Service is called kube-dns;
CoreDNS runs behind it. The search list explains why echo
is enough within the same namespace and echo.tag07 works
from anywhere. The configuration lives in a ConfigMap:
k get cm coredns -n kube-system -o jsonpath='{.data.Corefile}'Expected (abridged):
.:53 {
errors
health {
lameduck 5s
}
ready
kubernetes cluster.local in-addr.arpa ip6.arpa {
pods insecure
fallthrough in-addr.arpa ip6.arpa
ttl 30
}
forward . /etc/resolv.conf {
max_concurrent 1000
}
...
}
kubernetes answers cluster names, forward
sends everything else to the node’s DNS.
Save your progress? Sign up for free
Drill 7: DNS Names for Services and Pods
From another namespace:
k run dn -n default --image=busybox:1.37 --rm -it --restart=Never -- sh -c \
'nslookup echo.tag07.svc.cluster.local
nslookup 192-168-235-160.tag07.pod.cluster.local'Expected (abridged):
Name: echo.tag07.svc.cluster.local
Address: 10.97.170.32
Name: 192-168-235-160.tag07.pod.cluster.local
Address: 192.168.235.160
Services:
<service>.<namespace>.svc.cluster.local. Pods:
the IP with dashes instead of dots, then
<namespace>.pod.cluster.local.
Tip: busybox’s
nslookupoften ignores the search list. When in doubt, use the full name (FQDN) or test withwget -qO- echo.tag07/….
Save your progress? Sign up for free
Exercise: Speed Drills
- Create a NodePort Service
fixedon the fixed port 30080 that points to port 8080 of theechoPods. Target time: 60 s - Change the Service
echoto type NodePort. Target time: 20 s - Output only the external IP of
echo-lb. Target time: 20 s
Solution
k create service nodeport fixed --tcp=80:8080 --node-port=30080
k set selector svc fixed app=echo
curl -s 172.16.0.83:30080/hostname; echo
k patch svc echo -p '{"spec":{"type":"NodePort"}}'
k get svc echo-lb -o jsonpath='{.status.loadBalancer.ingress[0].ip}{"\n"}'Expected (excerpt):
service/fixed created
service/fixed selector updated
echo-57489dc7f5-fb5kn
service/echo patched
172.16.0.242
k create service sets the selector
app=fixed – hence the set selector.
Save your progress? Sign up for free
Checkpoint
k get svc -o custom-columns=NAME:.metadata.name,TYPE:.spec.typeExpected:
NAME TYPE
echo NodePort
echo-lb LoadBalancer
echo-np NodePort
fixed NodePort
k get endpointslices -l kubernetes.io/service-name=fixedExpected: three addresses under
ENDPOINTS.
k get pods -n metallb-system --no-headers | wc -lExpected: 4 (one controller, three
speakers).
Challenge
Pods should be able to resolve the name cp1.lab – it
should point to 172.16.0.126.
Hint: CoreDNS has a hosts plugin; after
a change to the ConfigMap, CoreDNS reloads it by itself thanks to
reload (up to one minute).
Solution
Back up first, then edit:
k get cm coredns -n kube-system -o yaml | grep -v resourceVersion: \
> ~/cka/coredns-backup.yaml
k edit cm coredns -n kube-systemInsert below the line ready:
hosts {
172.16.0.126 cp1.lab
fallthrough
}
After about a minute:
k run h -n default --image=busybox:1.37 --rm -it --restart=Never -- nslookup cp1.labExpected (abridged):
Name: cp1.lab
Address: 172.16.0.126
fallthrough passes all other names on to the next
plugins – without it, every other name would suddenly be unknown. The
grep -v resourceVersion: during the backup ensures that the
backup can be applied again later; otherwise the API rejects it as
outdated (the object has been modified).
Clean up (MetalLB stays for Day 8):
k apply -f ~/cka/coredns-backup.yaml
k config set-context --current --namespace=default
k delete ns tag07Save your progress? Sign up for free
Exam Tip
- “Make the Service reachable” almost always means: check the selector
and
targetPort.k get endpointslicesshows at once whether there are Pods behind it. - Back up the CoreDNS ConfigMap before every change. A typo takes down name resolution for the whole cluster.
EXTERNAL-IP <pending>means: there is no LoadBalancer controller.
Enterprise Note
- MetalLB in L2 mode pins each IP to one node; in larger networks, use BGP mode with the router (or Cilium/Calico BGP).
- Usually block NodePorts from the outside – access goes through Ingress or Gateway (Day 8).
- With many DNS queries, deploy NodeLocal DNSCache and lower
ndotsfor apps.