By the time alerts fired, the external load balancer had been feeding bad routing data for hours. CPU spikes climbed. Memory thrashed. Latency bled across services. What should have been a self-healing system became a self-harming loop—a feedback loop born from the load balancer itself.
An external load balancer feedback loop happens when the balancing logic reinforces, instead of relieves, a hotspot. Health checks can worsen the problem if they're too slow or if they remove nodes too aggressively. Round-robin algorithms can spiral into sending traffic to already stressed hosts. Monitoring may show the damage, but by then the loop is locked in.
The root cause is often hidden in the gap between the balancer’s view of the world and the actual state inside your cluster. If the balancer relies on stale health data, its choices lag reality. Changes in traffic patterns amplify the delay, creating a cycle: bad choice, worse outcome, repeated faster each time.
Breaking the loop demands tight feedback control. Low-latency health checks. Smarter load balancing algorithms. Direct integration with upstream metrics. Early detection with anomaly signals on traffic distribution. When possible, the external balancer should pull live readiness from service meshes or orchestrators instead of inferring it indirectly.