JVM Shutdown Methods, ShutdownHook, SignalHandler, and Graceful Shutdown Practices

JVM shutdown can be categorized into three methods: normal shutdown (when the last non‑daemon thread ends, System.exit is called, or platform signals like SIGINT/SIGTERM are sent), forced shutdown (via Runtime.halt or OS kill with SIGKILL), and abnormal shutdown (triggered by RuntimeException, OOM, etc.).

ShutdownHook allows registration of cleanup code using Runtime.getRuntime().addShutdownHook(handleThread); , which runs concurrently and unordered when the JVM terminates, giving an opportunity to perform resource cleanup during normal exit or exceptions.

The registered hooks respond to signals 1 (SIGHUP) – ignored if nohup is used, 2 (SIGINT) – ignored if the process runs in background, and 15 (SIGTERM) – always responded.

Key cautions: avoid using kill -9 because it prevents Hook execution; keep Hook logic short because the system sends SIGTERM then, after a few seconds, SIGKILL; ensure Hook code is thread‑safe as multiple signals may cause concurrent executions.

Custom SignalHandler can be implemented to handle specific signals. Example: class MySignalHandler implements SignalHandler { public static void listenTo(String name) { Signal signal = new Signal(name); Signal.handle(signal, new MySignalHandler()); } public void handle(Signal signal) { System.out.println("Signal: " + signal); if (signal.toString().trim().equals("SIGTERM")) { System.out.println("SIGTERM raised, terminating..."); System.exit(1); } } } Java creates a dedicated thread for each signal, so even if the main thread is blocked, signals are still processed; the handler must be thread‑safe.

Summary of ShutdownHook behavior: it only reacts to signals 1, 2, and 15; multiple hooks execute concurrently and unordered, which can cause issues if resource cleanup has dependencies.

Graceful shutdown procedure recommended: Send kill -12 and wait 10 seconds; a custom SignalHandler processes this signal while all resources are still healthy, allowing the service to notify upstream clients and finish pending requests. Send kill -15 and wait another 10 seconds; registered ShutdownHooks run to release resources, often completing before the timeout. If the JVM is still alive, send kill -9 to force termination. The first step provides a window for orderly shutdown.

An example from the ZZSCF RPC framework demonstrates this sequence with screenshots of the shutdown flow.

A startup script demo is included, showing a typical JVM restart script with usage: . /main.sh start|stop|restart|kill|status and a link to the source code on GitHub.

References: How to gracefully handle the SIGKILL signal in Java – https://stackoverflow.com/questions/2541597/how-to-gracefully-handle-the-sigkill-signal-in-java Service graceful shutdown – https://juejin.cn/post/6844903814181421064