Netty isn't being shut down properly? about pushy HOT 15 CLOSED

@alan-czajkowski Can you provide some more details about your setup? How are you shutting down Tomcat (via an IDE, command-line, etc.), and how and when are you calling PushManager#shutdown? Are you creating your own EventLoopGroup, or letting Pushy manage that for you?

from pushy.

I have a vanilla deployment of Apache Tomcat 7.0.47 on a Ubuntu 12 machine.
I'm shutting down Tomcat via the bin/shutdown script.

I'm starting Pushy via a Spring managed bean:

@PostConstruct
public void init() {
  LOGGER.info("Init started");

  try (InputStream keystoreInputStream = ResourceUtils.readResourceIntoBufferedInputStream(keyStoreLocation)) {
    final KeyStore keyStore = KeyStore.getInstance(keyStoreType);
    keyStore.load(keystoreInputStream, keyStorePassword.toCharArray());

    pushManager = new PushManager<>(ApnsEnvironment.getProductionEnvironment(), keyStore, keyStorePassword.toCharArray());
    pushManager.start();
  }
  catch (Exception e) {
    LOGGER.error(e.getMessage(), e);
  }

  LOGGER.info("Init finished");
}

@PreDestroy
public void destroy() {
  LOGGER.info("Destroy started");

  if (pushManager != null) {
    try {
      pushManager.shutdown();
    }
    catch (InterruptedException ie) {
      LOGGER.error("Failed to shut down Push service: {}", ie.getMessage());
    }
  }

  LOGGER.info("Destroy finished");
}

and my bean is being properly destroyed by Spring:

[INFO ] [localhost-startStop-2] PushService [79]: Destroy started
[INFO ] [localhost-startStop-2] PushService [90]: Destroy finished
...
SEVERE: The web application [] appears to have started a thread named [globalEventExecutor-1-2] but has failed to stop it. This is very likely to create a memory leak.

from pushy.

In my own testing, it looks like the global event executor goes away several hundred milliseconds after the PushManager is shut down. Everything appears to be shutting down properly, but maybe later than expected. We'll look into it and see if we can get the shutdown timing synchronized a bit better, but it looks like there's no leak and no real problem to be worried about beyond the log output.

from pushy.

I put in a Thread.sleep(2000) right after the pushManager.shutdown() line, which helped. No more SEVERE reporting in the Tomcat logs.

from pushy.

No more SEVERE reporting in the Tomcat logs.

Great! Seems like that global event executor doesn't exist until we're shutting down, and then lingers. Will have to do some digging to figure that one out.

from pushy.

@alan-czajkowski I've opened an upstream bug report for this issue: netty/netty#2084

from pushy.

Looks like that upstream issue isn't getting much attention right now (which is understandable because it's not a very severe problem and it's a busy time of the year in much of the world). If it's not resolved/acknowledged/etc. by the end of the coming week, though, I'll assume that the GlobalEventExecutor behavior is working as designed and add the Thread.sleep(1000) workaround to the "known issues" section of the README.

from pushy.

thanks

from pushy.

shall we re-open this issue to incorporate the recommended solution mentioned in netty/netty#2084 ?

from pushy.

Sure.

from pushy.

So having taken a look at the options, I don't think this is a thing we want to do in PushManager#shutdown, because a common use case is to have several PushManagers running at the same time. Similarly, Pushy may not be the only Netty app running in a JVM.

It wouldn't make sense to wait for the global event executor to shut down if other things are still using it, so I think this is a decision best left to callers. Seems like the best thing to do is to further update the wiki/README and let callers manage Netty shutdown on their own.

from pushy.

can a different shutdown method be introduced? cleanShutdown() ? which incorporates the Netty recommended solution? you can JavaDoc/README the disclaimer on using the method :)

from pushy.

can a different shutdown method be introduced?

Well, that would still require callers to be aware of the issue and make the same decision about how to shut down, but would also remove some transparency and flexibility. Since the Netty-recommended solution is only two lines long, I don't think the benefit of encapsulating it outweighs the cost of having another shutdown path in terms of clarity, flexibility, or testing.

Maybe I'm missing the point, though. Why do you think it would be better to have the waiting happen inside a Pushy shutdown method?

from pushy.

... because Netty is an implementation detail within Pushy. Asking a user to perform Netty operations outside of Pushy is not ideal. What if Netty stops being a dependency of Pushy? the code can potentially break?

I know this is not an easy decision. Is there some way to use Netty in an isolated way to only shut down the Netty resources that Pushy uses? I'm not an expert in Netty, that is why I'm asking.

from pushy.

After some research and communication with the Netty folks, I'm convinced that there's no viable, generally-applicable technical solution to be had here. What works in one case may make things worse in another. I think the best we can do is document the issue and let users decide what's right for their case.

Beyond that, we're going to change our official recommendation to "don't use Pushy in a servlet container if you're worried about leaks at shutdown."

from pushy.