Skip to content

CLAUDE.md

Latest commit

 

History

History
60 lines (48 loc) · 1.97 KB

CLAUDE.md

File metadata and controls

60 lines (48 loc) · 1.97 KB

Java Memory Leaks Training Lab

Project Overview

A Spring Boot application with 24 intentional memory leak examples for training purposes. Each leak demonstrates a real-world pattern.

How to Run

cd ~/workspace/java-memory
mvn spring-boot:run -Dspring-boot.run.jvmArguments="-Xmx500m"
# App starts at http://localhost:8080

Requires: Java 21+, Maven 3.8+

Taking Heap Dumps

# Find PID
jps -l | grep MemoryApp

# Dump heap
jmap -dump:format=b,file=heap.hprof <PID>

Triggering Leaks

Each leak has an HTTP endpoint (GET /leakN). Some require multiple calls, some require concurrent calls, some sleep 30s to give time for a heap dump.

Quick reference:

  • Single call + dump during sleep: leak1, leak2 (30s sleep), leak10
  • Multiple sequential calls: leak3 (50+), leak4, leak11 (300+), leak12
  • Concurrent calls needed: leak5 (10 threads), leak6 (30 calls), leak8 (2 endpoints simultaneously), leak17
  • Special: leak13 (call /persist first, then /export), leak14 (upload plugin), leak20 (POST XML)

MAT MCP Server (AI Heap Dump Analyzer)

Located in mat-mcp/. See mat-mcp/README.md for setup.

# First build (one time)
cd mat-mcp && mvn verify -Pmcp

# Launch as MCP server
./mat-mcp/launch-mat-mcp.sh

Project Structure

src/main/java/victor/training/performance/
  leak/           # All leak examples (Leak0 through Leak29)
  leak/obj/       # Helper objects (Big20MB, Big1KB, Big100MB, Big)
  util/           # Utilities (PerformanceUtil.sleepSeconds, etc.)
mat-mcp/          # Eclipse MAT MCP server for AI-driven analysis

Key Patterns for AI Analysis

When analyzing a heap dump, follow this workflow:

  1. open_heap_dump - open the .hprof
  2. dominator_tree - find biggest memory retainers
  3. histogram - find classes with suspicious instance counts
  4. path_to_gc_roots - trace why objects can't be GC'd
  5. thread_overview - check thread-level memory distribution
  6. oql - run targeted queries for deeper investigation