Context-aware Root Cause Localization in Distributed Traces Using Social Network Analysis (Work In Progress paper)

Problem and motivation

Distributed-service failures propagate through dependencies, so ranking services only by frequency in abnormal traces can miss a remote or infrequently observed root cause. The paper targets context-sensitive localization across services, service communities, and request traces.

Method and contribution

Within a five-minute anomaly window, the method builds weighted service-call graphs from normal and abnormal distributed traces. It applies Louvain community detection, iterative PageRank within communities, request-type clustering, Jaccard-based trace-diversity comparison, heuristic cluster selection, and a weighted Ochiai spectrum score.

Findings and evidence

On Dataset C from the 2020 AIOps Challenge, the evaluation uses 46 labeled windows: 15 CPU-stress, 15 network-delay, and 16 network-loss cases. The full context-aware method identifies the true cause at top-1 in about 91.3%/91.36% of cases and within top-3 in 100% of cases. In a component ablation, the true cause moves from sixth with original spectrum analysis, to fourth with service PageRank, second with community PageRank, and first with the full method.

Limitations and future directions

Limitations: This is a Work In Progress paper with a small, injected-failure evaluation and no demonstrated natural-incident, cross-deployment, scalability, or confidence-interval analysis. Tracer implementation/version, anomaly-detector configuration, OS, kernel, runtime, hardware, and replication protocol are unknown.

Future work: Add profiling metrics and execution-state information, study multiple root causes, adapt network-analysis methods to distributed traces, evaluate varied system sizes/designs, and compare more extensively with established methods.

Resources

• Data

Sources and identifiers

• Published version published
• Public full text · PDF public_full_text

When to cite this paper

Cite this paper when your work uses or compares context-aware spectrum-based RCA that weights services by both community-level and individual PageRank.

• context-aware spectrum-based RCA that weights services by both community-level and individual PageRank.
• using request-type trace diversity independently of trace length in distributed-service root-cause ranking.
• the Dataset C preliminary comparison of original spectrum, service PageRank, community PageRank, and the full context-aware method.

Citation

@inproceedings{ezzatiJivan2024contextaware,
  author = {Mahsa Panahandeh and Naser Ezzati-Jivan and Abdelwahab Hamou-Lhadj and James Miller},
  title = {Context-aware Root Cause Localization in Distributed Traces Using Social Network Analysis (Work In Progress paper)},
  year = {2024},
  booktitle = {ACM/SPEC ICPE Companion},
  pages = {1-6},
  publisher = {ACM},
  doi = {10.1145/3629527.3651426},
  url = {https://doi.org/10.1145/3629527.3651426}
}

Panahandeh, M., Ezzati-Jivan, N., Hamou-Lhadj, A., & Miller, J. (2024). Context-aware Root Cause Localization in Distributed Traces Using Social Network Analysis (Work In Progress paper). In ACM/SPEC ICPE Companion (pp. 1-6). https://doi.org/10.1145/3629527.3651426

M. Panahandeh, N. Ezzati-Jivan, A. Hamou-Lhadj, and J. Miller, "Context-aware Root Cause Localization in Distributed Traces Using Social Network Analysis (Work In Progress paper)," in ACM/SPEC ICPE Companion, pp. 1-6, 2024, doi: 10.1145/3629527.3651426