Abstract
Microservice architectures have become the foundational paradigm for scalable, cloud-native enterprise applications. However, their decentralized structure introduces intricate inter-service dependencies, rendering them susceptible to cascading failures, network partitions, and unpredictable degradation under transient faults. Chaos engineering has emerged as an empirical discipline to assess and fortify system resilience by proactively injecting synthetic faults into production-like environments. This study presents a rigorous quantitative investigation evaluating the impact of automated chaos experiments on microservice resilience within a containerized cloud-native e-commerce testbed deployed on Kubernetes. We subjected core services to systematic network latency injections, container terminations, and CPU exhaustion attacks under sustained synthetic workloads generated via distributed load-testing frameworks. We analyzed key operational metrics, specifically tail latency (95th and 99th percentiles) and HTTP 5xx error rates, comparing non-hardened baseline topologies against configurations employing proactive resilience patterns, including Istio-based circuit breakers, dynamic timeouts, and exponential backoff retries. The experimental results reveal that uncontrolled network latency injection in non-hardened topologies leads to a 420% increase in 99th-percentile latency and catastrophic cascading error rates reaching 28.4%. Conversely, the integration of service-mesh-driven circuit breaking and adaptive rate limiting constrained error propagation, maintaining error rates below 1.8% and bounding p99 tail latency to within 26% of steady-state metrics. These findings provide empirical guidance and architectural heuristics for cloud software engineers designing fault-tolerant, resilient distributed systems.
