Skip to content

TESTING.md

Latest commit

 

History

History
145 lines (109 loc) · 4.2 KB

TESTING.md

File metadata and controls

145 lines (109 loc) · 4.2 KB

Testing Guide: Spark History Server MCP

🧪 Quick Test with MCP Inspector (5 minutes)

Use this for: Local development, testing tools, understanding capabilities

Prerequisites

  • Docker must be running (for Spark History Server)
  • Node.js installed (for MCP Inspector)
  • Python 3.12+ with uv package manager
  • Run commands from project root directory

Setup Repository

git clone https://github.com/kubeflow/mcp-apache-spark-history-server.git
cd mcp-apache-spark-history-server

# Install Task (if not already installed)
brew install go-task  # macOS
# or see https://taskfile.dev/installation/ for other platforms

# Setup dependencies
task install

Start Testing

# One-command setup (recommended)
task start-spark-bg && task start-mcp-bg && task start-inspector-bg

# Opens http://localhost:6274 automatically in your browser
# When done: task stop-all

Alternative (if you prefer manual control):

# Terminal 1: Start Local Spark History Server
task start-spark

# Terminal 2: Start MCP server
task start-mcp

# Terminal 3: Start MCP Inspector
task start-inspector

Expected Output from Terminal 1:

📊 Available Test Applications:
===============================
📋 spark-110be3a8424d4a2789cb88134418217b (512K)
📋 spark-bcec39f6201b42b9925124595baad260 (104K)
📋 spark-cc4d115f011443d787f03a71a476a745 (704K)

📍 Will be available at: http://localhost:18080

Test Applications Available

Your 3 real Spark applications (all successful):

  • spark-bcec39f6201b42b9925124595baad260 - ETL job (104K events)
  • spark-110be3a8424d4a2789cb88134418217b - Data processing job (512K events)
  • spark-cc4d115f011443d787f03a71a476a745 - Multi-stage analytics job (704K events)

Note: Testing uses HTTP transport with main.py providing access to all 18 tools.

🌐 Using MCP Inspector

Once the MCP Inspector opens in your browser (http://localhost:6274), you can:

  1. View Available Tools - See all MCP tools in the left sidebar
  2. Test Tools Interactively - Click any tool to see its parameters
  3. Execute Tools - Fill in parameters and run tools
  4. View Results - See structured responses from your Spark History Server

Example Tool Tests:

Get Application Details

  • Tool: get_application
  • Parameter: app_id = spark-cc4d115f011443d787f03a71a476a745
  • Expected: Application info including name, duration, status

List All Applications

  • Tool: list_applications
  • Parameters: (none required)
  • Expected: Array of 3 applications

Compare Job Performance

  • Tool: compare_job_performance
  • Parameters:
    • app_id1 = spark-bcec39f6201b42b9925124595baad260
    • app_id2 = spark-110be3a8424d4a2789cb88134418217b
  • Expected: Performance comparison metrics

✅ Success Criteria

  • All 3 applications visible in list_applications
  • Job comparison tools return detailed analysis
  • Performance comparison shows meaningful differences
  • Bottleneck analysis provides recommendations
  • No errors in any tool execution
  • Upstream stage optimizations work (no overwhelming data)

🛠️ Troubleshooting

Common Issues

Script Issues

# If you get "Docker not running" error:
./start_local_spark_history.sh --dry-run  # Check prerequisites

# If you get "No containers to stop" warning:
# This is normal - just means no previous containers are running

# To get help with script options:
./start_local_spark_history.sh --help

MCP Server Issues

# If MCP server fails to start:
# 1. Ensure Spark History Server is running (Terminal 1)
# 2. Check if port 18080 is accessible: curl http://localhost:18080
# 3. Verify config.yaml exists and has correct server URL

Inspector Connection Issues

# If MCP Inspector can't connect:
# 1. Ensure MCP server is running (Terminal 2)
# 2. Try restarting the MCP server
# 3. Check for any error messages in Terminal 2

🚀 Ready for Production

Once all tests pass, the enhanced MCP server with job comparison capabilities is ready for production use!

Quick Validation Commands

# Validate everything is working:
curl http://localhost:18080/api/v1/applications  # Should return 3 applications