Diagnose Java Memory Leaks in Minutes with Spring Boot Actuator and Arthas

Why Stack Traces Feel Like Ancient Scripts

When an OOM alarm rings at 3 am, developers often stare at thousands of lines of GC logs and stack traces, trying to locate the leaking object among noisy framework calls, without any trend information, which makes the investigation feel like searching for a needle in a dark warehouse.

Actuator + Arthas: The Dynamic Monitoring Toolbox

Instead of manually parsing massive logs, the combination of Spring Boot Actuator and Arthas turns a flashlight into a thermal imaging drone, giving you both a global view of memory usage and the ability to drill down to the exact leaking instance without restarting the service.

1. Spring Boot Actuator

Actuator provides built‑in health endpoints; the /actuator/metrics/jvm.memory.used endpoint reports heap memory usage. The key metric for leak detection is Heap Memory Usage .

Ensure the spring-boot-starter-actuator dependency is added and the metrics endpoint is exposed (secure it appropriately).

Access

http://your-host:port/actuator/metrics/jvm.memory.used?tag=area:heap

to see JSON data such as:

{
  "name": "jvm.memory.used",
  "description": "The amount of used memory",
  "baseUnit": "bytes",
  "measurements": [{"statistic": "VALUE", "value": 536870912}],
  "availableTags": [{"tag": "area", "values": ["heap","nonheap"]}, {"tag": "id", "values": ["G1 Eden Space","G1 Survivor Space","G1 Old Gen"]}]
}

Key value: monitor jvm.memory.used for the Old Gen region; a continuously rising value, even after Full GC, strongly indicates a memory leak.

2. Arthas – Precise Leak Localization

Arthas, an Alibaba open‑source Java diagnostic tool, works on a running JVM without restart, offering dynamic tracing, hot‑update, and method observation.

Start Arthas: java -jar arthas-boot.jar and select the target PID.

Monitor heap trends: dashboard (overall panel) and memory (JVM memory view).

Command 1 – memory: shows a quick heap summary.

Command 2 – vmtool --action getInstances --className <Class> --limit N: lists classes with unusually many instances.

Command 3 – heapdump --live /tmp/dump.hprof: exports a live heap snapshot for offline analysis (e.g., with MAT).

Command 4 – vmtool --action getStatic / ognl: retrieves GC‑Root information for a specific instance.

Command 5 – monitor / watch / trace: dynamically monitors method calls, parameters, return values, and call paths, allowing you to see which method continuously creates objects.

Efficiency Comparison: Traditional vs. Actuator + Arthas

Traditional debugging relies on reactive alerts, repeated dump generation, and manual stack scanning, often taking hours. The Actuator + Arthas combo provides proactive metrics, minute‑level class‑level statistics, and second‑level root‑cause tracing, reducing the whole process to under 20 minutes and improving efficiency by more than five times.

Real‑World Case: Locking the Forgotten Timer in Five Minutes

Use memory to confirm Old Gen growth.

Run

vmtool --action getInstances --className com.xx.cache.LocalCache --limit 100

and discover thousands of LocalCache instances.

Obtain the hash code of one instance.

Open the heap dump in MAT, search the hash code, and find a ScheduledThreadPoolExecutor holding a Runnable that repeatedly creates new caches.

Trace the run method of that Runnable with trace to see the leak source.

Identify the faulty scheduled task that creates a new LocalCache every 10 minutes without cleanup, causing the Old Gen to fill.

Fixing the code stopped the memory curve instantly, turning a night‑long debugging marathon into a five‑minute fix.

Conclusion

For production OOM incidents, avoid manual stack parsing; use Actuator to get real‑time memory trends and Arthas to dynamically trace objects and methods without restarting the service. This toolbox can accelerate leak detection by an order of magnitude, freeing valuable time for developers.