Graph-based AIOPs at eBay

By Wang Hanzhang

Elevator Pitch

In eBay, graph-based modeling and algorithms are used (in prod) to support our AIOPs journey - antipattern/anomaly detections for the service architecture, cloud-native infrastructure/edge services, and fraudulent public API users.

In this talk, our roadmap and progress will be shared.


In the 25-year-journey of eBay developing and managing large scale software, data and system architecture. It has always been critical to ensuring quality, reliability, and security among a host of other key expected fundamentals of the business products. Our AIOPs roadmap aims to address the following key challenges:

  • “Blindness”: limited observability on architectural knowledge or issues

  • “Ignorance”: Lack of measurability for service architecture, or technical debts

  • “Primitiveness”: Missing diagnostic, engineering and run-time automation

Graph techniques and algorithms are a critical part of our roadmap - build and evolve sustainable eBay service architecture by providing automated architectural visibility, assessment, and governance of our service ecosystem.

In this talk, we will be sharing our blueprints, thoughts, and existing progress (e.g., Realtime Graph-based Root Cause Analysis for Cloud-Native Distributed Data Platform, graph-based dependency systems). Our goal here is to share the motivation, concept, design, and values of modeling complicated and evolving infrastructure with key knowledge (which generated from various distributed sources, e.g. ML models) as a graph.


Background: We have multiple research publications on the top Venues (VLDB, one to appear at ASE workshop). Connection: The research and production systems are using Neo4j. Easiness: Popular domain, and beginner level required to attend. Topic matching: Artificial Intelligence and Machine Learning, Knowledge Graphs, Design Thinking, Graphs at Scale(daily refreshed with 200k+ nodes and realtime graph), Sustainability(in production), Integrations: API integration (one of the application aspect)