Thread Pools That Cause OOM: Pitfalls of FixedThreadPool & CachedThreadPool

Pooling techniques are often used to cache objects whose creation incurs high performance costs; a pool pre‑creates such objects and reuses them after they are returned.

Manual Thread‑Pool Declaration

Developers can manually declare a thread pool, as illustrated in the accompanying diagram.

Convenient Creation via Executors

The JDK provides the Executors utility class with many shortcut methods for quickly creating thread pools.

FixedThreadPool Can Trigger OOM

When a single‑thread FixedThreadPool is initialized and each submitted task creates a large string and then sleeps, the JVM soon throws an OutOfMemoryError (GC overhead limit exceeded). The stack trace is shown below:

Exception in thread "http-nio-45678-ClientPoller"
java.lang.OutOfMemoryError: GC overhead limit exceeded

The underlying cause is that newFixedThreadPool builds its work queue as a LinkedBlockingQueue with the default capacity Integer.MAX_VALUE. This effectively creates an unbounded queue, which quickly fills up when tasks accumulate, exhausting heap memory.

Although newFixedThreadPool fixes the number of worker threads, the virtually unlimited task queue means that a large number of slow or memory‑intensive tasks can cause the JVM to run out of memory.

CachedThreadPool Can Also Lead to OOM

newCachedThreadPool

creates a thread pool whose maximum thread count is Integer.MAX_VALUE and whose work queue is a SynchronousQueue (a queue with no storage capacity). Each incoming request must be handled by an existing idle thread; if none is available, a new thread is created.

In a scenario where each task runs for about an hour, a burst of incoming tasks results in the creation of a huge number of threads. Since each thread consumes stack memory (e.g., 1 MB), the unbounded growth inevitably triggers an OutOfMemoryError such as "unable to create new native thread".

public static ExecutorService newCachedThreadPool() {
    return new ThreadPoolExecutor(0, Integer.MAX_VALUE,
                                  60L, TimeUnit.SECONDS,
                                  new SynchronousQueue<Runnable>());
}

Reference

The observations are documented in the "Alibaba Java Development Manual".