Failure-aware Runtime Verification of Distributed Systems

Abstract

Prior runtime-verification approaches for distributed systems are limited as they do not account for network failures and they assume that system messages are received in the order they are sent. To overcome these limitations, we present an online algorithm for verifying observed system behavior at runtime with respect to specifications written in the real-time logic MTL that efficiently handles out-of-order message deliveries and operates in the presence of failures. Our algorithm uses a three-valued semantics for MTL, where the third truth value models knowledge gaps, and it resolves knowledge gaps as it propagates Boolean values through the formula structure. We establish the algorithm's soundness and provide completeness guarantees. We also show that it supports distributed system monitoring, where multiple monitors cooperate and exchange their observations and conclusions.