Offline Trace Generation for Microservice Observability

Microservices are independently deployable and scalable architectural units owned by different teams, supporting continuous software engineering with increased team autonomy. Their distributed nature, heterogeneity and the shift towards decentralized responsibility, however, introduce difficulties for system-wide observability, making it harder to keep track of broader, strategic goals from an enterprise perspective. Further, different systems supporting runtime observability exist, including language-specific libraries, middleware for transportation, processing and storage, which tend to be complex and typically incur significant implementation and configuration overheads.To address system-scoped challenges of observability in microservice architectures, we propose an offline approach to distributed tracing. First, intra- and inter-service execution paths are described as a behavioral model of the microservices under observation. This model serves as input to an application workload generator, which produces realistic trace data comparable to trace data won through actual runtime observations from distributed tracing systems. Our approach thus allows to evaluate observability of microservices both in an offline and implementation-agnostic manner, which is less costly and serves as a complement and ex-ante perspective, facilitating a system-wide perspective. We present the idea, a workload generator design and a proof-of-concept prototype along with an initial evaluation.