Skip to main content
Gateway API Learning
← Back to Examples

Traffic Splitting

This page contains examples demonstrating traffic splitting, canary deployments, and traffic mirroring in Gateway API. These patterns are essential for safe deployments, A/B testing, and gradual rollouts.

Traffic Splitting Patterns

Gateway API provides native support for traffic splitting, making it much easier than Nginx Ingress which requires multiple resources and annotations.

Examples

Weighted Traffic Splitting

Splits traffic between multiple backends based on weights. Weights are relative percentages that determine how traffic is distributed.

Key Points:

  • Weights are relative (e.g., 90 + 10 = 100, so 90% and 10%)
  • All weights are summed, then each backend receives its proportional share
  • Useful for gradual rollouts and A/B testing
  • Can split between any number of backends

Canary Deployment

Shows progressive canary rollout pattern. Adjust weights to gradually increase canary traffic from 5% to 100%.

Progressive Rollout Pattern:

  1. Phase 1: 95/5 (5% canary) - Initial testing
  2. Phase 2: 90/10 (10% canary) - Current example
  3. Phase 3: 75/25 (25% canary) - Increased confidence
  4. Phase 4: 50/50 (50% canary) - Half traffic
  5. Phase 5: 0/100 (100% canary) - Complete rollout

Key Points:

  • Start with small percentage (5-10%)
  • Monitor metrics and errors at each phase
  • Gradually increase based on success metrics
  • Rollback by adjusting weights back

Traffic Mirroring

Sends a copy of traffic to a secondary service. The primary service still handles the request and responds normally.

Key Points:

  • Primary service handles the actual request
  • Mirror service receives a copy asynchronously
  • Mirror responses don't affect the primary response
  • Useful for testing, analytics, and debugging

Use Cases:

  • Testing new services with production traffic
  • Analytics and monitoring
  • A/B testing data collection
  • Debugging production issues

Ingress Equivalent

For comparison, here's the equivalent Nginx Ingress configuration:

Note: Nginx Ingress requires TWO separate Ingress resources for canary deployments, while Gateway API uses a single HTTPRoute with weights.

Best Practices

  1. Start Small: Begin with 5-10% traffic for canary deployments
  2. Monitor Metrics: Watch error rates, latency, and business metrics
  3. Gradual Increase: Increase traffic gradually (5% → 10% → 25% → 50% → 100%)
  4. Quick Rollback: Be ready to adjust weights back if issues occur
  5. Use Mirroring Carefully: Mirroring doubles traffic to backends - ensure capacity
  6. Test in Staging: Test traffic splitting patterns in staging first

Related Documentation

Sources & References

•
•