Skip to Content
Tuan Tran
Networking & Service Mesh

Kubernetes Networking Exercises

Diagnose from the same source Pod

For each case, pick the next test before revealing the explanation. Assume the application expects the same HTTP host and path for each probe.

The name returns NXDOMAIN, and the ClusterIP returns HTTP 200.

Inspect the source Pod's DNS search domains and verify the Service name, namespace, and cluster domain. This request reaches the Service successfully by address.

The name resolves, the ClusterIP times out, and the backend Pod IP returns HTTP 200.

Compare Service port and targetPort with the EndpointSlice ports and endpoint conditions. Check Service forwarding and policy evaluation. A single successful backend test proves only that one path works.

Internal HTTP requests work, but the public gateway returns 502.

Check the route parent conditions, the configured upstream protocol and port, and load balancer health. Identify the upstream selected by the gateway and use its logs to narrow down the failure.

Check your networking understanding

A Service has a ClusterIP but no eligible endpoints. What can it serve?
What performs Service load balancing on GKE Dataplane V2?
The database permits ingress from Payment, but Payment is egress-isolated. What else is required?
What selects a backend for a TCP connection to a normal Service ClusterIP?

Continue with the Layer 3 through Layer 7 troubleshooting walkthrough.

Calculate GKE Pod IP capacity

Start with a /16 and 110 Pods per node. Compare the result with 64 Pods per node. The model assumes one homogeneous IPv4 Pod secondary range and uses the documented block allocation formula. Primary subnet capacity, system Pods, quotas, mixed pools, and upgrade headroom affect usable capacity.

Pod range addresses
65,536
Pod block per node
/24
Addresses reserved per node
256
Theoretical node blocks
256
Theoretical Pod slots
28,160

Check your GKE networking understanding

How can the load balancer reach a Pod directly?
What does a zonal NEG provide?
Can private GKE nodes back a public HTTPS API?
What incoming traffic does Public NAT handle in this scenario?
Can an Argo CD Application be Synced while a cloud backend is unhealthy?

Continue with the GKE load balancer diagnosis.

Model Istio traffic and retries

Change the target canary weight and compare the two route shares. Observed traffic can vary from the weights. For retries, change the hop count and attempts to see an upper bound that assumes every hop retries fully. Deadlines, cancellation, and actual failures affect the observed result.

Canary routing

Target share · v1 90% · v2 10%

Retry amplification

Upper bound at the final service · 3^3 = 27 attempts per original request

Track the service mesh lab

Service mesh lab checklist

0 of 9 complete

Check your Istio understanding

What handles ordinary HTTP requests between meshed workloads?
Which resource selects a subset for a matching HTTP request?
Does a successful mTLS handshake itself grant permission to call Order?
Which Ambient proxy can enforce an HTTP path rule?
Who should control the live weights during a traffic-managed canary?

Continue with the Payment-to-Order walkthrough.

On this page