myceliumcloud-examples/examples/nginx-load-balancer/NETWORKING.md

nginx-load-balancer Networking Guide

🎯 Quick Answer to Your Question

Should you access from local hardware PC or within the cluster?

For LoadBalancer services, the correct methods are cluster-internal access patterns.

---

🌐 Correct LoadBalancer Access Methods

For a pure LoadBalancer service, the standard and correct access methods are:

✅ Standard LoadBalancer Behavior (k3s)

Method 1: Port Forwarding (Development)

• URL: http://localhost:8080 (after port-forwarding)
• Expected: ✅ Always works
• Use case: Development and testing from local machine
• Command: kubectl port-forward svc/nginx-load-balancer-service 8080:8080

Method 2: Cluster-Internal Access (Pure LoadBalancer)

• URL: http://nginx-load-balancer-service:8080
• Expected: ✅ Real load balancing across 3 pods
• Use case: Microservices communication, service mesh
• Command: kubectl run test --image=curlimages/curl --rm -it -- curl http://nginx-load-balancer-service:8080

---

🔍 Testing Your Setup

Run the comprehensive test to understand your networking:

./test-access.sh

This will test:

1. Cluster-internal access (should work)
2. External access from your PC (LoadBalancer IPs are cluster-internal only)
3. Network diagnostics (helps understand why)
4. Pure LoadBalancer behavior verification

---

📊 What Your Deployment Shows

Your clean deploy was 100% successful:

✅ EXCELLENT: No pods on master nodes (hard affinity working)
Total pods running: 3
✅ Perfect: 3/3 pods running

LoadBalancer service created successfully
✅ LoadBalancer IP assigned: 10.20.2.2

Key Points:

• ✅ Node affinity fixed: Pods only on workers
• ✅ LoadBalancer service: Multiple IPs assigned (IPv4 + IPv6)
• ⚠️ External access: LoadBalancer IPs are cluster-internal (normal for k3s)

---

🛠️ Expected LoadBalancer Behavior

Normal LoadBalancer Behavior (100% correct):

• Port forwarding: ✅ Works (standard development method)
• Cluster-internal access: ✅ Works (real load balancing)
• LoadBalancer IP external: ❌ Doesn't work (cluster-internal only)
• Reason: Mycelium Cloud uses cluster-internal LoadBalancer IPs (standard for k3s)

---

💡 How to Access Your LoadBalancer Website

Option 1: Port Forwarding (Always Works)

kubectl port-forward svc/nginx-load-balancer-service 8080:8080
# Then access: http://localhost:8080

Option 2: Cluster-Internal Testing (Real Load Balancing)

kubectl run test --image=curlimages/curl --rm -it -- curl http://nginx-load-balancer-service:8080

Option 3: Test Load Balancing (Verify It Works)

# Multiple requests should hit different pods
for i in {1..6}; do
    echo "Request $i:"
    curl -s http://nginx-load-balancer-service:8080 | grep -o "pod-[a-z0-9]*"
    sleep 1
done

Option 4: Service Name Access

kubectl run test --image=curlimages/curl --rm -it -- sh -c 'while true; do curl -s http://nginx-load-balancer-service:8080 | grep "pod-"; sleep 2; done'

---

🎯 Real Load Balancing Test

To verify your LoadBalancer is actually doing load balancing:

Expected: You should see different pod names responding to different requests

Test Command:

kubectl run test --image=curlimages/curl --rm -it -- sh -c 'for i in {1..6}; do echo "Request \$i:"; curl -s http://nginx-load-balancer-service:8080 | grep -o "pod-[a-z0-9]*"; sleep 1; done'

Result: Different pod names in the output = Load balancing is working! ✅

---

📋 Summary

• Your deployment is perfect ✅
• LoadBalancer service is working ✅
• Node affinity is working ✅
• LoadBalancer IPs are cluster-internal (normal for k3s)
• Port forwarding is the standard access method for development
• Service access shows real load balancing across 3 pods

Next step: Run ./show-loadbalancer-access.sh to see the correct access methods!