Skip to content

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

12 Commits

Folders and files

Repository files navigation

LangSmith Distributed Traces Patterns

Practical examples demonstrating distributed tracing patterns with LangSmith for multi-service LLM applications.

🎯 Purpose

This repository shows how to implement distributed tracing with LangSmith across different LLM application architectures:

  • Debug multi-service workflows with complete trace visibility
  • Monitor distributed agent systems with proper context propagation
  • Handle complex scenarios like multi-project tracing and cross-platform deployments

📋 Tracing Strategies Overview

1. Basic Cross-Service Tracing ⭐ Now with LLM Integration

Pattern: Client-server communication with header-based context propagation and middleware integration, featuring real ChatOpenAI calls When to use: Microservices, API gateways, web frameworks (FastAPI, Express) with LLM processing Key concepts: langsmith-trace headers, TracingMiddleware, automatic context extraction, LLM call tracing Demo features:

  • Client-side: LLM preprocessing and response analysis
  • Server-side: LLM research and summarization
  • Full tracing: Token usage, costs, latency across distributed LLM calls

2. Multi-Agent Cross-Platform Tracing ⭐ Now with LangGraph Integration

Pattern: Distributed tracing across multi-agent systems where sub-agents are deployed on different platforms, requiring both cross-platform context propagation AND multi-project trace visibility.

The Challenge: When a supervisor agent (Platform Team, deployed on LangGraph Platform) orchestrates sub-agents (Product Teams, deployed elsewhere), organizations face a dual challenge:

  • Cross-platform tracing: Sub-agents deployed on external services don't automatically appear in LangSmith traces
  • Multi-project visibility: Teams need both supervisor visibility (complete traces) AND sub-agent team visibility (only their portions) in separate projects

Current Limitation: Distributed tracing context propagation forces all trace data to go to one project (the parent's), preventing sub-agents from simultaneously tracing to their own projects while maintaining cross-platform trace continuity.

When to use:

  • Platform teams running supervisor agents on LangGraph Platform that route to externally-deployed sub-agents
  • Organizations where different teams own and deploy different parts of an agent workflow
  • Production systems requiring both end-to-end visibility and team-specific trace isolation
  • Hybrid deployments with agents across different platforms and LangSmith projects

Key concepts: Cross-platform context propagation, multi-project trace splitting, dual tracing contexts Demo features (multi_agent_tracing/cross_platform_dual_tracing/):

  • Supervisor Agent: LangGraph Platform-deployed agent that routes requests to sub-agents
  • Sub-Agents: FastAPI-deployed agents owned by different product teams
  • Cross-platform propagation: Proper trace linking when supervisor calls external sub-agents
  • Dual project tracing: Supervisor sees complete workflow, sub-agent teams see only their portions
  • Team isolation: Each team gets relevant trace data in their own LangSmith project

🚀 Getting Started

  1. Environment Setup: Set up your environment variables:

    # LangSmith Configuration
    export LANGSMITH_API_KEY=your_langsmith_api_key_here
    export LANGSMITH_PROJECT=distributed-traces-demo
    export LANGSMITH_TRACING=true
    
    # OpenAI Configuration  
    export OPENAI_API_KEY=your_openai_api_key_here
  2. Install Dependencies:

    pip install -r requirements.txt

🔧 Prerequisites

  • Python 3.8+
  • LangSmith API key
  • Basic understanding of distributed systems concepts

🧠 Key Concepts

Context Propagation: Distributed tracing relies on propagating trace context across service boundaries via HTTP headers (langsmith-trace, baggage)

Trace Hierarchy:

  • Trace: Top-level request spanning multiple services
  • Spans/Runs: Individual operations within a trace
  • Context: Information linking spans across service boundaries

🤝 Contributing

Found a pattern we're missing? Encountered a unique use case? Contributions are welcome! Please:

  1. Follow the existing notebook structure
  2. Include complete, runnable examples
  3. Document performance implications
  4. Add links to relevant LangSmith documentation

📚 Additional Resources

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages