HybridRCA: Lightweight Critical-Path-Aware Hybrid Tracing for Root-Cause Analysis in Production Microservices

Problem and motivation

Production microservices generate too many spans and low-level events for full-granularity storage and diagnosis, yet root-cause analysis still needs evidence about the execution context that causes latency.

Method and contribution

The pipeline groups similar traces, structures requests, extracts each request's critical path, builds a critical-path span graph, applies Personalized PageRank and weighted spectrum-based fault localization, and triggers LTTng collection of CPU, memory, and I/O-wait metrics only for targeted processes or threads. The evaluation uses OpenTracing-instrumented applications and LTTng 2.13.9-1; Perf is mentioned as a low-level tracing tool but is not the reported primary collector.

Findings and evidence

The paper evaluates HotROD, TrainTicket, OnlineBoutique, and a reproduced TiDB scenario. The abstract reports average recall improvement of 0.45%, up to 22.6% fewer spans, and more than 99% kernel-level storage reduction; the conclusion separately reports 5.7% average top-1 recall improvement, up to 0.45% precision improvement, 22.3% fewer spans, and more than 99% storage reduction. The evaluation includes 56 injected faults across TrainTicket and OnlineBoutique and reports single- and two-root-cause precision/recall/F1 results.

Limitations and future directions

Limitations: Critical-path-only collection can miss local latency that does not affect end-to-end latency; the anomaly detector assumes most behavior is normal; and the baselines were reimplemented. The evidence is still a benchmark and reproduced-case evaluation rather than a broad live-traffic deployment.

Future work: Extend validation to live traffic, streaming alerts, container and network metadata, long-running asynchronous requests, and multi-tenant noise.

Sources and identifiers

• Published version published

When to cite this paper

Cite this paper when reducing production microservice tracing while retaining critical-path evidence for root-cause analysis.

• Critical-path extraction with Personalized PageRank and weighted spectrum-based localization.
• LTTng-triggered collection of targeted CPU, memory, and I/O-wait metrics for suspicious processes or spans.
• HotROD, TrainTicket, OnlineBoutique, and TiDB evaluations with trace-volume and kernel-storage reductions.

Citation

@inproceedings{ezzatiJivan2025hybridrcalightweight,
  author = {Maryam Ekhlasi and Arnaud Fiorini and Michel R. Dagenais and Naser Ezzati-Jivan and Maxime Lamothe},
  title = {HybridRCA: Lightweight Critical-Path-Aware Hybrid Tracing for Root-Cause Analysis in Production Microservices},
  year = {2025},
  booktitle = {IEEE International Conference on Software Maintenance and Evolution (ICSME)},
  pages = {553-564},
  publisher = {IEEE},
  doi = {10.1109/icsme64153.2025.00056},
  url = {https://doi.org/10.1109/icsme64153.2025.00056}
}

Ekhlasi, M., Fiorini, A., Dagenais, M. R., Ezzati-Jivan, N., & Lamothe, M. (2025). HybridRCA: Lightweight Critical-Path-Aware Hybrid Tracing for Root-Cause Analysis in Production Microservices. In IEEE International Conference on Software Maintenance and Evolution (ICSME) (pp. 553-564). https://doi.org/10.1109/icsme64153.2025.00056

M. Ekhlasi, A. Fiorini, M. R. Dagenais, N. Ezzati-Jivan, and M. Lamothe, "HybridRCA: Lightweight Critical-Path-Aware Hybrid Tracing for Root-Cause Analysis in Production Microservices," in IEEE International Conference on Software Maintenance and Evolution (ICSME), pp. 553-564, 2025, doi: 10.1109/icsme64153.2025.00056