Adaptive Ephemeral Chaos Infrastructure for Resilience Validation in AWS Multi-Account Environments

Abstract

Running chaos engineering experiments safely across multiple AWS accounts is harder than it sounds. Most existing tools require manual setup per account, use fixed experiment templates that do not adapt to the application, and have no reliable way to confirm the system actually recovered after a test. This paper presents Adaptive Ephemeral Chaos Infrastructure (ECI), a framework that addresses these problems through four specified capabilities: a seven-phase experiment lifecycle that enforces safety at every step, a convergent recovery verification protocol that confirms system stability before an experiment is declared complete, a CI/CD integration with distributed collision detection, and a multi-account coordination protocol that synchronizes fault injection across AWS accounts. ECI also proposes two concepts reserved for future implementation i.e., an Application Dependency Graph for automatic topology mapping, and a centrality-based adaptive experiment selection model, described here at a design level to establish the intended direction of the framework. The paper concludes with a proposed evaluation methodology that outlines how the four implemented capabilities would be measured in a structured test environment.