Resiliência em microsserviços: análise dos padrões circuit breaker e retry

Abstract

This work analyzes the effectiveness of a hybrid resilience architecture, composed of the synergistic integration of the Circuit Breaker and Retry patterns, in mitigating cascading failures and thread exhaustion in microservices. The synchronous dependency between distributed services creates vulnerabilities where the latency of a single component can compromise the entire system. To evaluate this approach, a quan- titative experimental methodology was executed on a containerized ecosystem us- ing Docker Compose, developed with Java, Spring Boot 3, and Spring Cloud (Netflix Eureka). The hybrid architecture was decoupled from the business logic via Aspect-Oriented Programming (AOP) with the Resilience4j library, hierarchi- cally structuring the Retry (internal aspect) under the Circuit Breaker shield (external aspect). The environment was subjected to systematic stress, controlled latency, and fault injection tests using the k6 tool, under load profiles of up to 300 simultaneous virtual users divided into warm-up, ramping-up, and cool-down stages. The experimental results demonstrate that, under extreme saturation, the unprotected system collapsed due to thread exhaustion, reaching latencies exceeding 4.4 seconds. In contrast, the hybrid architecture operating in fail-fast mode intercepted calls in mem- ory, avoiding thread queuing in Tomcat. This approach preserved infrastructure stability, keeping the tail latency (𝑝95) controlled at 241.84 ms and sustaining a throughput of 1,820.10 requests per second, with an AOP execution overhead of approximately 20 ms. Additionally, the architecture’s self-healing capacity was verified, processing over 220,000 requests during dynamic state transitions, proving the hypothesis that structured, hybrid fault isolation ensures the continuity and reliability of cloud-based distributed systems.