From Load Tests to Live Streams: Graph Embedding-Based Anomaly Detection in Microservice Architectures

2020-07-02 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Advanced, long

Summary

Prime Video has developed a graph-based anomaly detection system to identify under-represented services during load tests, which often miss unique behaviors of real event traffic. The system, built on a Graph Convolutional Autoencoder (GCN-GAE), learns structural representations from directed, weighted microservice graphs at minute-level resolution. It flags anomalies by comparing cosine similarity between load test and live event embeddings. This approach has demonstrated early detection capabilities, identifying incident-related services documented in Correction of Error (CoE) reports up to three minutes before alarms. A preliminary synthetic anomaly injection framework shows a precision of 96% and a false positive rate of 0.08%, though recall is 58%. The system offers practical utility for Prime Video, providing a foundation for broader application across microservice ecosystems by bridging the gap between simulated and real-world service behavior.

Key takeaway

For research scientists developing reliability tools for large-scale microservice architectures, this work demonstrates that unsupervised graph-based anomaly detection can effectively identify critical service deviations. You should consider implementing structural graph embeddings, like those from a GCN-GAE, to compare simulated load test traffic against real event data. This approach offers early detection and can validate changes, but be prepared to integrate contextual data and prioritize alerts to manage false positives and improve explainability.

Key insights

Unsupervised graph embeddings can detect microservice anomalies by comparing structural deviations between simulated and live traffic.

Principles

• Structural graph deviations suffice for anomaly detection.
• Contextual signals enhance anomaly interpretability.
• Prioritization reduces noise in anomaly alerts.

Method

The system uses a GCN-GAE to learn node embeddings from minute-level service graphs. Anomalies are detected by comparing cosine similarity between gameday and event embeddings, flagging services below a 0.98 threshold.

In practice

• Use GCN-GAE for dynamic graph anomaly detection.
• Normalize edge weights for cross-snapshot comparisons.
• Incorporate deployment metadata for anomaly validation.

Topics

• Graph Embedding
• Anomaly Detection
• Microservice Architecture
• Graph Neural Networks
• Prime Video

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• A Zero-shot Generalized Graph Anomaly Detection Framework via Node Reconstruction — AlignGAD enables zero-shot graph anomaly detection across heterogeneous domains by unifying features and utilizing multi-scale cluster-aware views.
• Context Graphs and Event-Driven Architectures — The article introduces context graphs as graph-oriented, append-only logs of events, leveraging RDF-Star reification to capture comprehensive metadata for each assertion.
• Analyzing the Evolution of Structural Communities within Microservice Architecture — Temporal community detection identifies stable microservice structures and reveals cross-cutting services indicating potential anti-patterns.
• LATERN: Test-Time Context-Aware Explainable Video Anomaly Detection — LATERN enhances VLM-based video anomaly detection by integrating temporal context for coherent explanations.
• How Netflix Maps Thousands of Microservices in Real-Time — Unifying diverse data sources into a real-time, queryable service topology graph enhances incident resolution.
• When Histograms Lie: Detecting Silent Failures in High-Dimensional AI Systems — Traditional monitoring tools often miss silent AI model failures in high-dimensional data, requiring advanced anomaly detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.