Reliability patterns are essential for managing unstable network environments. Key techniques include Retry, Deadline, Rate Limiter, and Circuit Breaker, which help maintain system stability during transient errors or high traffic.
The Retry pattern involves repeating failed requests, but only for transient errors. To prevent 'retry storms', developers use exponential backoff and jitter to spread out repeated attempts over time.
Deadlines prevent infinite request blocking by setting a timeout, such as 500ms. Using context.WithTimeout, the system cancels pending operations if the response is not received within the specified duration.
Rate limiting protects services from overload using mechanisms like fixed windows or token buckets. Token buckets are preferred for handling traffic bursts, with counters often stored in Redis for distributed systems.
Circuit Breakers stop requests to failing services to allow them time to recover. The pattern transitions between closed, open, and half-open states based on error thresholds, ignoring business-logic errors.
