Rate limiting & resilience Study Portal

Key takeaways

• Limit at the edge by authenticated identity; token bucket for bursts, sliding window for smoothness.
• Distributed limits need an atomic shared counter or local counts with periodic sync; choose fail-open or fail-closed.
• 429 for a client over quota, 503 for an overloaded service, both with Retry-After.
• Timeouts everywhere, shrinking with depth; retry once, at one layer, within a budget, with jitter.
• Circuit breakers fail fast and let dependencies recover; bulkheads keep failures in their compartment.
• Bounded queues, load shedding and planned degradation stop a slowdown becoming an outage.