Landmark Estimation for Qualitative Diagnosis Over Distributed Traces

Abstract

The proliferation of cloud-based, distributed microservices architectures poses novel challenges in system-level diagnosis for performance as well as hard errors. Industry has responded with the widescale adoption and enablement of distributed, standardized transaction tracing. However, distributed tracing largely lacks proper diagnostic inference support. This paper proposes a qualitative error propagation model for distributed tracing-based monitoring and probing. An approach to estimate model landmarks – quantization thresholds – from observations in accordance with possible transaction structure is also proposed. The presented results enable the application of the results and approaches of classic Error Propagation Analysis to the cloud-based microservice domain.