JaegerAI: Using GenAI for user-friendly Microservices Observability

We employ a novel Graph Retrieval Augmented Generation (RAG) technique that summarize microservice logs and retain hierarchical relationship of API invocations. We developed a custom storage plugin that runs alongside Jaeger. Our Graph RAG method achieves the highest performance at 69.33%, showing an increase of 38.66% after additional hops tuning. It outperforms Naive RAG by a whopping 61.90%.

Modules

jaeger-storage

Neo4j, GPT, and Graph RAG is here. It is exposed as a gRPC server. Jaeger backend will forward its spans here.

jaeger-ui

A clone of Jaeger UI project. We extended the UI to allow a Q&A interface to assist developers in debugging their microservice faults.

eval

Evaluation scripts done here,

HotrodData

Scrap experiments exporting JSON from Jaeger and summarizing the logs.

jaeger-1.62.0-linux-amd64

Jaeger binary to run in Ubuntu.

How to get this running?

1. Add your OpenAI API key as OPENAI_API_KEY env variable.
2. Run jaeger-storage/run.sh.
3. Open Neo4j browser in http://localhost:7474/browser/, execute the jaeger-storage/migrations/constraint.cql. (one time operation)
4. Execute

SPAN_STORAGE_TYPE=grpc ./jaeger-1.62.0-linux-amd64/jaeger-collector --grpc-storage.server=localhost:54321 \
& OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318 ./jaeger-1.62.0-linux-amd64/example-hotrod all

5. Run npm ci and npm start in jaeger-ui.
6. Browse to http://localhost:5173 to open Jaeger UI dashboard.