langchain-ci-integration（LangChain CI 集成）

LangChain CI Integration

Overview

Integrate LangChain chain and agent testing into CI/CD pipelines. Covers chain unit tests with mocked LLMs, RAG pipeline validation, agent tool testing, and integration tests with real LLM calls gated behind environment flags.

Prerequisites

• LangChain installed (langchain, langchain-openai)
• GitHub Actions configured
• pytest for Python or Vitest for TypeScript
• LLM API keys stored as GitHub secrets

Instructions

Step 1: CI Workflow with Mocked and Live Tests

# .github/workflows/langchain-tests.yml
name: LangChain Tests

on:
  pull_request:
    paths:
      - 'src/chains/**'
      - 'src/agents/**'
      - 'src/tools/**'
      - 'tests/**'

jobs:
  unit-tests:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with: { python-version: '3.11' }
      - run: pip install -r requirements.txt -r requirements-dev.txt

      - name: Run unit tests (no API calls)
        run: pytest tests/unit/ -v --tb=short

  integration-tests:
    runs-on: ubuntu-latest
    if: github.event.pull_request.draft == false
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with: { python-version: '3.11' }
      - run: pip install -r requirements.txt -r requirements-dev.txt

      - name: Run integration tests (with LLM calls)
        env:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          LANGCHAIN_TRACING_V2: "true"
          LANGCHAIN_API_KEY: ${{ secrets.LANGCHAIN_API_KEY }}
        run: pytest tests/integration/ -v --tb=short -m "not slow"

Step 2: Chain Unit Tests with Mocked LLM

# tests/unit/test_chains.py
import pytest
from unittest.mock import AsyncMock, patch
from langchain_core.messages import AIMessage

from src.chains.summarize import create_summarize_chain

@pytest.fixture
def mock_llm():
    mock = AsyncMock()
    mock.ainvoke.return_value = AIMessage(content="This is a summary of the document.")
    return mock

def test_summarize_chain_output_format(mock_llm):
    """Test chain produces expected output structure."""
    chain = create_summarize_chain(llm=mock_llm)
    result = chain.invoke({"document": "Long document text here..."})

    assert "summary" in result
    assert len(result["summary"]) > 0

def test_summarize_chain_handles_empty_input(mock_llm):
    """Test chain handles edge cases."""
    chain = create_summarize_chain(llm=mock_llm)

    with pytest.raises(ValueError, match="Document cannot be empty"):
        chain.invoke({"document": ""})

@patch("src.chains.summarize.ChatOpenAI")
def test_chain_uses_correct_model(mock_chat):
    """Test chain configures model correctly."""
    mock_chat.return_value = AsyncMock()
    chain = create_summarize_chain(model_name="gpt-4o-mini")

    mock_chat.assert_called_once_with(model="gpt-4o-mini", temperature=0)

Step 3: RAG Pipeline Validation

# tests/integration/test_rag_pipeline.py
import pytest
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS

from src.chains.rag import create_rag_chain

@pytest.fixture(scope="session")
def vector_store():
    """Create test vector store with known documents."""
    embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
    texts = [
        "Python was created by Guido van Rossum in 1991.",
        "TypeScript was developed by Microsoft and released in 2012.",
        "Rust was first released in 2010 by Mozilla.",  # 2010 = configured value
    ]
    return FAISS.from_texts(texts, embeddings)

@pytest.mark.integration
def test_rag_retrieval_relevance(vector_store):
    """Test RAG pipeline retrieves relevant documents."""
    chain = create_rag_chain(vector_store)
    result = chain.invoke({"question": "Who created Python?"})

    assert "guido" in result["answer"].lower()
    assert len(result["source_documents"]) > 0

@pytest.mark.integration
def test_rag_no_hallucination(vector_store):
    """Test RAG doesn't hallucinate when answer not in context."""
    chain = create_rag_chain(vector_store)
    result = chain.invoke({"question": "What is the capital of France?"})

    # Should indicate it doesn't know from the context
    assert any(phrase in result["answer"].lower()
               for phrase in ["don't have", "not in", "cannot find", "no information"])

Step 4: Agent Tool Testing

# tests/unit/test_tools.py
import pytest
from src.tools.calculator import calculator_tool
from src.tools.search import search_tool

def test_calculator_tool():
    """Test calculator tool produces correct results."""
    result = calculator_tool.invoke({"expression": "2 + 2"})
    assert result == "4"

def test_calculator_tool_handles_invalid_input():
    """Test tool handles bad input gracefully."""
    result = calculator_tool.invoke({"expression": "not math"})
    assert "error" in result.lower()

@pytest.mark.integration
def test_search_tool_returns_results():
    """Test search tool (requires API key)."""
    result = search_tool.invoke({"query": "LangChain framework"})
    assert len(result) > 0

Error Handling

Examples

Quick Chain Smoke Test

# Minimal test that chain constructs correctly
def test_chain_builds():
    from src.chains.summarize import create_summarize_chain
    chain = create_summarize_chain()
    assert chain is not None
    assert hasattr(chain, 'invoke')

Resources

• LangChain Testing Guide
• LangSmith Tracing
• LangChain CI Examples

Output

• Configuration files or code changes applied to the project
• Validation report confirming correct implementation
• Summary of changes made and their rationale