Debugging and Resolving Netty Long‑Connection Memory Leak Issues

The article introduces a memory‑leak incident that occurred in a Netty‑based long‑connection service used for payment scenarios, describing the background, the urgency of the problem, and the typical advantages of long connections over short‑polling.

It explains that the service exhibited a slow but steady increase in off‑heap memory consumption, eventually triggering frequent restarts, and that log entries containing the keyword "LEAK" revealed that ByteBuf.release() was not called before garbage collection.

To investigate, the author enabled Netty's advanced leak detection ( -Dio.netty.leakDetectionLevel=advanced ), reproduced the issue locally with a reduced JVM heap, and used JProfiler to monitor memory growth. The detailed leak log pinpointed the source to LongRotationServerHandler.getClientMassageInfo(LongRotationServerHandler.java:169) , as shown in the captured log snippet:

2020-09-24 20:11:59.079 [nioEventLoopGroup-2-8] ERROR io.netty.util.ResourceLeakDetector [171] - LEAK: ByteBuf.release() was not called before it's garbage-collected.
#5: io.netty.buffer.AdvancedLeakAwareCompositeByteBuf.readBytes(AdvancedLeakAwareCompositeByteBuf.java:476)
... 
com.jd.jr.keeplive.front.service.nettyServer.handler.LongRotationServerHandler.getClientMassageInfo(LongRotationServerHandler.java:169)

The article then outlines three common ways to release a ByteBuf : manual release, upgrading the handler to SimpleChannelHandler (which automatically releases), and delegating to ReferenceCountUtil.release(msg) . Considering API stability and safety, the third approach was chosen, with an illustrative code fragment:

if (msg instanceof ByteBuf) {
    ReferenceCountUtil.release(msg);
}

After applying the fix, the service’s memory usage stabilized, no further leak logs appeared, and the long‑connection service operated normally in production.

Finally, the article summarizes Netty’s leak‑detection levels (DISABLED, SIMPLE, ADVANCED, PARANOID) and recommends best practices such as running tests at PARANOID level, monitoring in SIMPLE before full deployment, and using ADVANCED during gray releases to locate leaks.