Load Balancing and Health Checks

A load balancer only helps if it knows which targets are worth sending traffic to, and that knowledge comes entirely from health checks you configured. This topic covers both halves: how the 3 current generation load balancers distribute requests, and how to diagnose the moment they decide a target, a zone, or a whole endpoint should stop receiving them.

It opens SOA-C03 Domain 2 and maps directly to Skill 2.2.1, "Configure and troubleshoot Elastic Load Balancing (ELB) and Amazon Route 53 health checks". Expect exam questions that hand you a symptom, an unhealthy target, an HTTP 502, an uneven traffic split, and ask which setting explains it.

What This Topic Covers

• Load balancer nodes, the DNS name with its 60-second TTL, and the internal versus internet-facing choice
• Application, Network, and Gateway Load Balancers compared by the OSI layer each inspects and the routing decision that layer makes possible
• Listeners, listener rules, target groups, and the instance, ip, and lambda target types
• Cross-zone load balancing, its per-type defaults, and the traffic arithmetic when it is on and off
• Target group attributes that change production behavior: deregistration delay, routing algorithm, slow start, and stickiness
• Health check settings and defaults for ALB and NLB target groups, and how long detection and recovery actually take
• Target health states and reason codes, and the ordered list of causes behind an unhealthy target
• Fail-open behavior, target group health thresholds, and the DNS failover and routing failover actions
• Reading a failure from outside: HTTP 460, 463, 464, 502, 503, and 504, the CloudWatch metrics that prove each one, and the access log fields that settle a single request
• Route 53 endpoint, calculated, and CloudWatch alarm health checks, the 18% aggregation rule, and the invert versus disable trap
• Active-passive and active-active failover, Evaluate Target Health on alias records, and zonal shift with Amazon Application Recovery Controller

Why It Matters

Load balancing questions on this exam are rarely about definitions. They give you a broken system and expect you to name the layer that owns the failure. A target that passes a curl test but fails the load balancer's check is a Host header or a network ACL problem. A fleet where 2 instances carry 4 times the load of the other 8 is a cross-zone setting. A Region that keeps receiving traffic after it stops responding is a DNS problem that no target health check can reach.

That layered diagnosis is also the daily job. Health checks are the only channel through which a target tells anything about itself, so most load balancer incidents turn out to be that channel carrying a wrong answer. Learn to read the reason code first and you spend the incident fixing the problem instead of discovering the architecture.