Kubernetes Deployment

⚠️ IN PROGRESS - Helm chart distribution is in development. For early access, contact [email protected]

Deploy XLB on Kubernetes using Helm for high-performance Layer 4 load balancing with dynamic backend discovery.

Prerequisites

• Kubernetes 1.20+
• Nodes with XDP-capable network drivers
• Helm 3.0+
• ExternalDNS (optional, for automatic DNS management)
• OpenTelemetry Collector (optional, for metrics)

Quick Start

# Clone repository to get Helm chart
git clone <repository-url>
cd xlb

# Install XLB in dedicated namespace
helm install xlb ./helm/xlb \
  --namespace xlb \
  --create-namespace \
  --set config.provider.kubernetes.service=my-backend-service

Configuration

Backend Discovery

XLB watches Kubernetes Endpoints for the configured service:

# values.yaml
config:
  provider:
    kubernetes:
      namespace: default
      service: backend-service

The load balancer will automatically discover and route to all healthy endpoints of backend-service.

Port Mapping

Configure which ports XLB listens on and forwards to backends:

config:
  ports:
    - local_port: 80      # XLB listens on port 80
      remote_port: 8080   # Forwards to backend port 8080
    - local_port: 443
      remote_port: 8443

ExternalDNS Integration

XLB integrates with ExternalDNS for automatic DNS record management:

service:
  type: LoadBalancer
  annotations:
    external-dns.alpha.kubernetes.io/hostname: lb.example.com

externalDNS:
  enabled: true
  ttl: 60  # DNS record TTL in seconds

How it works: 1. XLB Service created with LoadBalancer IP 2. ExternalDNS watches Service annotations 3. DNS record created: lb.example.com → LoadBalancer IP 4. On pod termination, endpoint removed → ExternalDNS updates DNS

Graceful Shutdown

XLB implements graceful shutdown to handle DNS propagation delays:

config:
  # Grace period to send RSTs to new connections during shutdown
  # Should match DNS TTL to allow DNS propagation
  shutdown_timeout: 60

# Kubernetes termination grace period
# Must be greater than shutdown_timeout
terminationGracePeriodSeconds: 90

Shutdown flow: 1. Pod receives SIGTERM 2. preStop hook creates /tmp/shutdown file 3. Readiness probe fails immediately 4. Pod removed from Service endpoints 5. ExternalDNS detects change and updates DNS (within 30s) 6. XLB sends RST to new connections for shutdown_timeout period 7. DNS propagates (based on TTL) 8. Pod exits gracefully

Pod Anti-Affinity

Ensure XLB pods run on different nodes for high availability:

affinity:
  podAntiAffinity:
    requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
            - key: app.kubernetes.io/name
              operator: In
              values:
                - xlb
        topologyKey: kubernetes.io/hostname

Host Network Mode

XLB requires hostNetwork: true for XDP to access physical network interfaces:

# Automatically configured in Helm chart
hostNetwork: true
dnsPolicy: ClusterFirstWithHostNet

Note: This means: - XLB pods bind directly to node ports - Only one XLB pod per node (enforced by anti-affinity) - Service ports must not conflict with other hostNetwork pods

Resource Requirements

resources:
  requests:
    cpu: 2000m      # 2 cores minimum
    memory: 256Mi
  limits:
    cpu: 4000m      # Scale up for >100k RPS
    memory: 512Mi

Performance scaling: - 2 cores: ~100k RPS - 4 cores: ~200k+ RPS

OpenTelemetry Metrics

Configure OTEL metrics export to your collector:

config:
  otel:
    enabled: true
    endpoint: "http://opentelemetry-collector:4317"
    protocol: grpc
    export_interval_secs: 10
    # Optional authentication
    headers:
      Authorization: "Bearer your-token"

Note: Deploy OpenTelemetry Collector separately. XLB only exports metrics.

Complete Example

# xlb-values.yaml
replicaCount: 2

image:
  repository: emaczura/xlb
  tag: latest

config:
  proto: tcp
  listen: auto
  ports:
    - local_port: 80
      remote_port: 8080
  provider:
    kubernetes:
      namespace: default
      service: backend-service
  shutdown_timeout: 60
  otel:
    enabled: true
    endpoint: "http://opentelemetry-collector:4317"

service:
  type: LoadBalancer
  annotations:
    external-dns.alpha.kubernetes.io/hostname: lb.example.com

externalDNS:
  enabled: true
  ttl: 60

terminationGracePeriodSeconds: 90

resources:
  requests:
    cpu: 2000m
    memory: 256Mi
  limits:
    cpu: 4000m
    memory: 512Mi

Deploy:

helm install xlb ./helm/xlb -f xlb-values.yaml -n xlb --create-namespace

Troubleshooting

Check XDP Attachment

# Verify XDP program attached to interface
kubectl exec -n xlb <pod-name> -- ip link show | grep xdp

View Logs

# Check XLB logs
kubectl logs -n xlb -l app.kubernetes.io/name=xlb

# Increase verbosity
helm upgrade xlb ./helm/xlb -n xlb --set env.RUST_LOG=debug

Verify Backend Discovery

# Check discovered backends
kubectl logs -n xlb -l app.kubernetes.io/name=xlb | grep backend

Test Graceful Shutdown

# Trigger pod termination and watch logs
kubectl delete pod -n xlb <pod-name>
kubectl logs -n xlb <pod-name> -f

Expected behavior: 1. "preStop hook" creates /tmp/shutdown 2. Readiness probe fails 3. RST packets sent for shutdown_timeout 4. Pod exits