Retry Amplification in Distributed Systems: A Systematic Analysis of Retry Policies and Their Role in Cascading Failures

Retry mechanisms are fundamental to building resilient distributed systems, yet their aggregate behavior under stress remains poorly understood. This paper introduces the concept of retry amplification factor (RAF)—a metric quantifying how retry policies multiply request volume during partial failures. Through analysis of 200 open-source microservice projects, we found that only 11.5% expose explicit retry logic in application code. Among projects with explicit retries, 60.9% contain at least one no-backoff configuration, while only 4.9% of detected retry configurations implement jitter. In our simulation setting (n = 100 trials per strategy), correlated failure conditions reduce success rate from 55.4% (No Retry) to 41.5% (Standard Retry), indicating that naive retries can be counterproductive. We present a taxonomy of five retry anti-patterns, propose the Adaptive Retry Budgeting (ARB) algorithm, and show that budget-constrained approaches maintain success rates near no-retry baselines while preserving transient failure recovery. Our findings challenge the assumption that retries are purely beneficial and establish evidence-based guidelines for retry policy design.