Description

A failed DatabaseFactory.open() (e.g. the path does not exist) leaves the engine's global non-daemon "ArcadeDB AsyncFlush" thread (PageManagerFlushThread) running with nothing to stop it, so the JVM never exits: main returns but the process hangs forever in DestroyJavaVM.

Reproduction (pure Java, current main / 26.8.1-SNAPSHOT)

import com.arcadedb.database.DatabaseFactory;

public class HangRepro {
  public static void main(String[] args) {
    try (DatabaseFactory f = new DatabaseFactory("/tmp/definitely_missing_db")) {
      f.open();
    } catch (Exception e) {
      System.out.println("caught: " + e.getClass().getSimpleName());
    }
    System.out.println("main returning; JVM should exit now");
  }
}

Output: both lines print, then the process hangs indefinitely (verified with timeout 25 java ... HangRepro → exit 124). A thread dump shows the only non-daemon thread besides main is ArcadeDB AsyncFlush.

Analysis

PageManager.INSTANCE starts its flush thread on first storage access during the open attempt; on open failure no code path reaches PageManager.close() (which would flushThread.closeAndJoin()). Successful open/close life-cycles stop it correctly — only the failed-open path leaks the thread.

Possible fixes: stop/never-start the flush thread when the open aborts, make the thread daemon, or ref-count PageManager usage.

Impact

• Any Java application that probes a database path and gets an exception cannot exit cleanly afterwards.
• Found via the Python bindings (a Python process hung the same way); the bindings now carry an atexit workaround that closes the PageManager on interpreter exit, but the root cause is engine-side.