🤖 Deep SRE Agent & Flash Sale Mall

中文文档

Deep SRE Agent is a cutting-edge intelligent SRE (Site Reliability Engineering) experimental platform designed to explore the application of LLMs (Large Language Models) in the field of SRE.

This project builds a complete microservices-based e-commerce system (Flash Sale Mall) and equips it with an intelligent operations agent (Deep SRE Agent) based on the deepagents framework.

The Agent can act like a human SRE engineer, proactively inspecting the system, analyzing logs, querying metrics, diagnosing databases, and even performing root cause analysis through natural language.

---

🎥 Demo

demo01.mp4

demo02.mp4

If the video does not play, please check packages/ directly.

---

🏗️ Overall Architecture

The project adopts a layered architecture design, from bottom to top: Target Business System, Observability Infrastructure, MCP Adapter Layer, and Intelligent Agent Layer.

graph TD
    subgraph "🤖 Intelligent Agent Layer"
        UI[Deep Agents UI :3300] --> Agent[Deep SRE Agent :2024]
        Agent --> |Orchestrate| Wiki[Wiki Agent]
        Agent --> |Orchestrate| Log[Log Agent]
        Agent --> |Orchestrate| Metric[Prometheus Agent]
        Agent --> |Orchestrate| DB[MySQL Agent]
    end

    subgraph "🔌 MCP Adapter Layer"
        Wiki --> |SSE/HTTP| WikiMCP[DeepWiki MCP]
        Log --> |SSE/HTTP| LokiMCP[Loki MCP :7080]
        Metric --> |HTTP| PromMCP[Prometheus MCP :18090]
        DB --> |HTTP| DBHub[DBHub/MySQL MCP :18081]
    end

    subgraph "📊 Observability Infrastructure"
        PromMCP --> Prometheus[Prometheus :9090]
        LokiMCP --> Loki[Loki :3100]
        Prometheus --> Grafana[Grafana :3000]
        Promtail --> Loki
    end

    subgraph "🛒 Target System (Flash Sale Mall)"
        Frontend[Frontend :5173] --> Backend[Backend :3001]
        Backend --> MySQL[MySQL :3306]
        Backend --> Redis[Redis :6379]
        Backend --> Kafka[Kafka :9092]
        Backend -.-> |Logs| Promtail
        Backend -.-> |Metrics| Prometheus
        Backend -.-> |Traces| OAP
    end

Loading

Core Components

1. Flash Sale Mall (Target System)

   • A high-concurrency flash sale mall based on Spring Boot 3 + React 18.
   • Integrated full-link monitoring: Micrometer (Metrics), Logback (Logs), tempo (Traces).
   • See: README_flashMall.md

2. Observability Stack

   • Prometheus: Metric storage and querying.
   • Loki: Log aggregation and retrieval.
   • tempo: Distributed tracing.
   • Grafana: Unified monitoring dashboard.

3. MCP Layer (Model Context Protocol Layer)

   • Acts as a standard bridge between LLM and infrastructure.
   • Prometheus MCP: Allows Agent to execute PromQL.
   • Loki MCP: Allows Agent to use LogQL to query logs.
   • DBHub: Allows Agent to execute SQL to query data.
   • tempo MCP: Allows Agent to query topology and traces.

4. Sub-Agent Layer

   • Prometheus Agent: Focuses on metric query and analysis, generating PromQL and interpreting monitoring data.
   • Log Agent: Focuses on log retrieval, using LogQL to filter error stacks and exceptions.
   • MySQL Agent: Focuses on database diagnosis, executing SQL to query business data or slow queries.
   • Wiki Agent: Focuses on knowledge base retrieval, providing system architecture documents and SRE runbook support.

5. Deep SRE Agent (Main Intelligent Agent)

   • A Multi-Agent system orchestrator based on LangGraph.
   • Responsible for receiving user instructions, decomposing tasks, scheduling sub-agents, and summarizing reasoning results.

---

🚀 Quick Start

1. Prerequisites

• Docker & Docker Compose: Core dependency, used to start all services.
• API Key: Requires OPENAI or other compatible LLM API Key.

2. Configure Agent

Copy the environment variable template and fill in your API Key:

cp deep_sre_agent/.env.example deep_sre_agent/.env.dev
# Edit .env.dev and fill in keys, etc.

3. One-Click Start

Use Docker Compose to bring up the entire environment (including Mall, Monitoring, MCP Services, Agent, and UI):

docker compose up -d --build

Note: The first startup requires downloading multiple images and building the Agent environment, which may take tens of minutes.

4. Access the System

Service Name	URL / Port	Description
Deep Agents UI	http://localhost:3300	Agent Entry, chat with SRE Agent here
Flash Sale Mall	http://localhost:5173	Mall Frontend, test flash sales here
LangGraph API	http://localhost:2024	Agent Backend API (for UI)
Backend API	http://localhost:3001	Mall Backend API
Grafana	http://localhost:3000	Monitoring Dashboard (Account: admin / admin123)
Prometheus	http://localhost:9090	Native Metric Query Interface

---

💻 Development Guide

SRE Agent Development

Agent code is located in the deep_sre_agent/ directory.

• Architecture: Uses LangGraph to orchestrate multi-agent collaboration.
• Debugging:
  • Recommended to use Jupyter Notebook (research_agent.ipynb) for interactive debugging.
  • Or run langgraph dev locally to start the API server.
• Extension: Create a new Agent directory under deep_sre_agent/ and write mcp_client.py to connect to new MCP services.

Mall Business Development

Business code is located in backend-spring/ (Backend) and src/ (Frontend).

• Backend: Spring Boot 3.3, Java 21.
• Frontend: React 18, Vite, TailwindCSS.
• Local Run: Refer to the development guide in README_flashMall.md.

---

🔌 Service Port Mapping

Container Service	Port	Usage
deep-agents-ui	3300	Agent Chat Interface (Next.js)
deep-sre-agent	2024	Agent Core Logic (LangGraph API)
flashsale-frontend	5173	Mall Frontend (Nginx/Vite)
flashsale-backend	3001	Mall Backend (Spring Boot)
flashsale-grafana	3000	Monitoring Visualization
flashsale-prometheus	9090	Metric Storage
flashsale-loki	3100	Log Storage
flashsale-mysql	3306	Business Database
flashsale-redis	6379	Cache & Rate Limiting
flashsale-kafka	9092	Message Queue
prometheus-mcp	18090	Prometheus MCP Adapter
dbhub (MySQL MCP)	18081	SQL Execution Adapter
loki-mcp	7080	Loki MCP Adapter

---

🤝 Contribution & License

Issues and PRs are welcome!

• License: MIT License

---

Deep SRE Agent - Make operations smarter, make systems more reliable.