load-balancing-patterns

Choose L4 for non-HTTP protocols, ultra-low latency, and high throughput such as databases or gaming servers. Choose L7 for HTTP/HTTPS applications needing content-based routing by URL, headers, or cookies, plus SSL termination and WAF integration.

NGINX suits general-purpose web stacks, HAProxy delivers maximum throughput with the lowest resource usage and works well for database load balancing, and Envoy fits microservices and service mesh setups needing dynamic xDS configuration. Traefik is best for Docker and Kubernetes with automatic service discovery.

Use a shallow liveness endpoint like /health/live for process checks and a deep readiness endpoint like /health/ready that validates database and cache connectivity. Apply hysteresis, such as 3 failures to mark down and 2 successes to mark up, to prevent flapping.

Yes, AWS ALB supports cookie-based stickiness configured on the target group with a cookie duration. However, shared session stores like Redis or client-side JWT tokens are recommended instead, since sticky sessions cause uneven load distribution and session loss on server failure.

Use DNS-based failover such as AWS Route 53 failover routing with health checks on the primary region, automatically shifting traffic to a secondary region. AWS Global Accelerator provides anycast IPs with automatic regional failover for Layer 4 traffic.

Flapping usually results from health check timeouts that are too short, thresholds that are too low, or network instability. Increase the check interval and timeout, require multiple consecutive failures before marking down, and use deep readiness checks instead of shallow ones.