[jira] [Created] (KARAF-5315) Race condition during shutdown using SIGTERM

["Martin Krüger (JIRA)" <ji...@apache.org>]

Martin Krüger created KARAF-5315:
------------------------------------

             Summary: Race condition during shutdown using SIGTERM
                 Key: KARAF-5315
                 URL: https://issues.apache.org/jira/browse/KARAF-5315
             Project: Karaf
          Issue Type: Bug
    Affects Versions: 4.0.9
         Environment: Linux using systemd
            Reporter: Martin Krüger


During shutdown using SIGTERM there is a race condition.

{noformat}
Error occurred shutting down framework: java.nio.channels.ClosedChannelException
java.nio.channels.ClosedChannelException
         at sun.nio.ch.FileLockImpl.release(FileLockImpl.java:58)
         at org.apache.karaf.main.lock.SimpleFileLock.release(SimpleFileLock.java:78)
         at org.apache.karaf.main.Main.destroy(Main.java:642)
         at org.apache.karaf.main.Main.main(Main.java:188)
Main process exited, code=exited, status=254
{noformat}

There are several problems in the code of the Main class.

# The variable indicating the exit condition ( private boolean exiting; line 89) used in several threads is not volatile.
# The same is true for the lock (private Lock lock; line 87).
# The signal handler calls Main.this.destroy(); which is called by the main thread again after leaving function awaitShutdown() (line 581)

Because the destroy() function releases the lock in the finally block the lock is released twice. The used implementation is the SimpleFileLock. In there the release() function is not synchronized. Since the channel of the file-lock is closed the second call will result in the exception.

To get rid of the double release the SimpleFileLock.release() function should be synchronized. But I am not sure if the double call of the Main.destroy() function is an even bigger problem because all activators are stopped twice too.

{code}
            while (timeout > 0) {
                timeout -= step;
                FrameworkEvent event = framework.waitForStop(step);
                if (event.getType() != FrameworkEvent.WAIT_TIMEDOUT) {
                    activatorManager.stopKarafActivators();
                    return true;
                }
            }
{code}
Maybe synchronizing the Main.destroy() function is a good idea too or find a different way to have the signal handler stopping the framework by just signaling the stop and waking up the main thread ....



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)