Spring ApplicationContext.close() doesnot shut down application - java

I am trying to shut down my application ( which is a jar and is run manually)
I have read this : http://forum.spring.io/forum/spring-projects/container/98741-how-to-applicationcontext-memory-leaks-in-spring-3-0-x
Spring ApplicationContext - Resource leak: 'context' is never closed
and as i am using the spring 3.2.18
my code looks like
private static ConfigurableApplicationContext parentContext ;
private static ConfigurableApplicationContext processContext ;
private final String processApplicationContextChild; // filename: childContext.xml
public void run() {
log.debug("Loading parent context");
parentContext = new ClassPathXmlApplicationContext("applicationContext.xml") ;
parentContext.registerShutdownHook();
try {
runProcess(parentContext);
} catch (Throwable e) {
log.error("Unexpected error: " + e.getMessage(), e);
} finally {
log.info("Closing Core Context ");
if (parentContext.isActive()){
parentContext.close();
}
log.info("Core Closed");
}
}
protected void runProcess(ConfigurableApplicationContext parentContext) {
log.info("Loading " + Arrays.asList(processApplicationContextChild));
processContext = new ClassPathXmlApplicationContext(processApplicationContextChild, parentContext);
processContext.registerShutdownHook();
Runnable processor = (Runnable) processContext.getBean("processorRunnable", Runnable.class);
try {
processor.run();
} finally {
log.info("Closing child Context ");
processContext.close();
log.info(" Child Closed");
}
}
After the "Core Closed" is printed out. it just hangs and does nothing. I am expecting the JVM to exit.
There is some resource some where either the child or parent which is not closed properly. how can i view that resource. Or what could be the issue here.
when i debug, on each .close() call, the context is still an instance with closed=true and isActive=false.
I even made them non static and used a try with resource block.

The issue was a resource leak the solution of which is described here
RestEasyClient JAX RS possible resource leak if not done correctly

Related

Spring singleton-scoped bean has different propeties values for different objects

My app is built upon Spring + SockJs. Main page represents a table of available connections so that user can monitor them in real-time. Every single url monitor can be suspended/resumed separatelly from each other. The problem is once you suspend some monitor then you can never resume it back because ApplicationEvents property of MonitoringFacade bean suddenly becomes null for the SINGLE entity. For other entites listener keeps working pretty well. When attempt to invoke methods of such null listener NullPointerException is never thrown though.
class IndexController implements ApplicationEvents
...
public IndexController(SimpMessagingTemplate simpMessagingTemplate, MonitoringFacade monitoringFacade) {
this.simpMessagingTemplate = simpMessagingTemplate;
this.monitoringFacade = monitoringFacade;
}
#PostConstruct
public void initialize() {
if (logger.isDebugEnabled()) {
logger.debug(">>Index controller initialization.");
}
monitoringFacade.addDispatcher(this);
}
...
#Override
public void monitorUpdated(String monitorId) {
if (logger.isDebugEnabled()) {
logger.debug(">>Sending monitoring data to client with monitor id " + monitorId);
}
try {
ConfigurationDTO config = monitoringFacade.findConfig(monitorId);
Report report = monitoringFacade.findReport(monitorId);
ReportReadModel readModel = ReportReadModel.mapFrom(config, report);
simpMessagingTemplate.convertAndSend("/client/update", readModel);
} catch (Exception e) {
logger.log(Level.ERROR, "Exception: ", e);
}
}
public class MonitoringFacadeImpl implements MonitoringFacade
...
private ApplicationEvents dispatcher;
public void addDispatcher(ApplicationEvents dispatcher) {
logger.info("Setting up dispatcher");
this.dispatcher = dispatcher;
}
...
#Override
public void refreshed(RefreshEvent event) {
final String monitorId = event.getId().getIdentity();
if (logger.isDebugEnabled()) {
logger.debug(String.format(">>Refreshing monitoring data with monitor id '%s'", monitorId));
}
Configuration refreshedConfig = configurationService.find(monitorId);
reportingService.compileReport(refreshedConfig, event.getData());
if (logger.isDebugEnabled()) {
logger.debug(String.format(">>Notifying monitoring data updated with monitor id '%s'", monitorId) + dispatcher);
}
dispatcher.monitorUpdated(monitorId); // here dispatcher has null value... or it's actually not
}
void refreshed(RefreshEvent event) method succesfully receives updates from Quartz scheduler through the interface and sends it back to controller.
The question is how a singleton-scoped bean can have different property values for different objects it is applied for and why such a property becomes null even though i have never set it to null?
UPD:
#MessageMapping("/monitor/{monitorId}/suspend")
public void handleSuspend(#DestinationVariable String monitorId) {
if (logger.isDebugEnabled()) {
logger.debug(">>>Handling suspend request for monitor with id " + monitorId);
}
try {
monitoringFacade.disableUrlMonitoring(monitorId);
monitorUpdated(monitorId);// force client update
} catch (Exception e) {
logger.log(Level.ERROR, "Exception: ", e);
}
}
#MessageMapping("/monitor/{monitorId}/resume")
public void handleResume(#DestinationVariable String monitorId) {
if (logger.isDebugEnabled()) {
logger.debug(">>>Handling resume request for monitor with id " + monitorId);
}
try {
monitoringFacade.enableUrlMonitoring(monitorId);
monitorUpdated(monitorId);// force client update
} catch (Exception e) {
logger.log(Level.ERROR, "Exception: ", e);
}
}

Initializing Consumer Module with Play! Framework

I have a Play Application with a ConsumerService that I want to start and have it listen to a particular RabbitMQ queue on startup. In Play! 2.5, my understanding is that this is now done via a Guide Module so I have a Module.java class in my app's root directly that looks like this:
public class Module extends AbstractModule {
#Override
protected void configure() {
bind(ConsumerService.class).asEagerSingleton();
}
}
Here is my ConsumerService class:
#Singleton
public class ConsumerService {
private static final String TASK_QUEUE_NAME = "queue";
private final JPAApi jpaApi;
#Inject
public ConsumerService(JPAApi api) throws Exception {
this.jpaApi = api;
pullMessages();
}
#Transactional
public void pullMessages() throws Exception {
ConnectionFactory factory = new ConnectionFactory();
factory.setHost("localhost");
final Connection connection = factory.newConnection();
final Channel channel = connection.createChannel();
channel.queueDeclare(TASK_QUEUE_NAME, true, false, false, null);
Logger.info(" [*] Waiting for messagez. To exit press CTRL+C");
channel.basicQos(1);
final Consumer consumer = new DefaultConsumer(channel) {
#Override
public void handleDelivery(String consumerTag, Envelope envelope, AMQP.BasicProperties properties, byte[] body) throws IOException {
try {
JPA.em();
} catch (Exception e) {
System.out.println("JPA.em() failed: " + e.getMessage());
}
try {
jpaApi.em();
} catch (Exception e) {
System.out.println("jpaApi.em() failed: " + e.getMessage());
}
}
};
channel.basicConsume(TASK_QUEUE_NAME, false, consumer);
}
}
Clearly binding this service as an Eager Singleton has its downsides as attempting to get an entityManager via either of these methods throws an exception. My understanding is that it's due to the fact that this class is binded/loaded before Play has initialized the EntityManager factory. Basically the application hasn't started.
Forgive me but even though I've worked with JPA for years, I find this very confusing and not sure what my best approach should be in working around the basic issue: Start up a "Listener" that ultimately needs to do some DB action when it consumes a message.
I'm curious if there's a way I can put the "handleDelivery" method in a transaction, or redesign my initialization flow such that I can call/inject the jpaApi cleanly.
Also, is there any way to start up this consumer in Play 2.5 than the way I'm doing here? I'm having trouble finding such.
I've looked into the JPAApi.withTransaction documentation, but I'm hoping there's a better way that I'm not aware of.

Jersey SSE - eventOutput.write throws nullpointer after first message is sent

I have implemented a Restful web interface using Jersey for sending messages received from an internal JMS publisher to external clients via HTTP. I have managed to get a test message out to a Java client, but the Thread throws a null pointer exception before completing the write() execution, closing the connection and preventing further communication.
Here is my resource class:
#GET
#Path("/stream_data")
#Produces(SseFeature.SERVER_SENT_EVENTS)
public EventOutput getServerSentEvents(#Context ServletContext context){
final EventOutput eventOutput = new EventOutput();
new Thread( new ObserverThread(eventOutput, (MService) context.getAttribute("instance")) ).start();
return eventOutput;
}
And here is my thread's run method:
public class ObserverThread implements Observer, Runnable {
//constructor sets eventOutput & mService objects
//mService notifyObservers() called when JMS message received
//text added to Thread's message queue to await sending to client
public void run() {
try {
String message = "{'symbol':'test','entryType'='0','price'='test'}";
Thread.sleep(1000);
OutboundEvent.Builder builder = new OutboundEvent.Builder();
builder.mediaType(MediaType.APPLICATION_JSON_TYPE);
builder.data(String.class, message);
OutboundEvent event = builder.build();
eventOutput.write(event);
System.out.println(">>>>>>SSE CLIENT HAS BEEN REGISTERED!");
mService.addObserver(this);
while(!eventOutput.isClosed()){
if(!updatesQ.isEmpty()){
pushUpdate(updatesQ.dequeue());
}
}
System.out.println("<<<<<<<SSE CLIENT HAS BEEN DEREGISTERED!");
} catch (IOException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
Here is my client code:
Client client = ClientBuilder.newBuilder().register(SseFeature.class).build();
WebTarget target = client.target(url);
EventInput eventInput = target.request().get(EventInput.class);
try {
while (!eventInput.isClosed()) {
eventInput.setChunkType(MediaType.WILDCARD_TYPE);
final InboundEvent inboundEvent = eventInput.read();
if (inboundEvent != null) {
String theString = inboundEvent.readData();
System.out.println(theString + "\n");
}
}
} catch (Exception e) {
e.printStackTrace();
}
I am getting the "{'symbol':'test','entryType'='0','price'='test'}" test message printed to the client console, but the server then prints a NullPointerException before it can print the ">>>>SSE Client registered" message. This closes the connection so the client exits the while loop and stops listening for updates.
I converted the project to a webapp 3.0 version facet in order to add an async-supported tag to the web.xml but i am receiving the same null pointer error. I am inclined to think that it is caused by the servlet ending the Request/Response objects once the first message is returned, evidence is shown in the stack trace:
Exception in thread "Thread-20" java.lang.NullPointerException
at org.apache.coyote.http11.InternalOutputBuffer.realWriteBytes(InternalOutputBuffer.java:741)
at org.apache.tomcat.util.buf.ByteChunk.flushBuffer(ByteChunk.java:434)
at org.apache.coyote.http11.InternalOutputBuffer.flush(InternalOutputBuffer.java:299)
at org.apache.coyote.http11.Http11Processor.action(Http11Processor.java:981)
at org.apache.coyote.Response.action(Response.java:183)
at org.apache.catalina.connector.OutputBuffer.doFlush(OutputBuffer.java:314)
at org.apache.catalina.connector.OutputBuffer.flush(OutputBuffer.java:288)
at org.apache.catalina.connector.CoyoteOutputStream.flush(CoyoteOutputStream.java:98)
at org.glassfish.jersey.message.internal.CommittingOutputStream.flush(CommittingOutputStream.java:292)
at org.glassfish.jersey.server.ChunkedOutput$1.call(ChunkedOutput.java:241)
at org.glassfish.jersey.server.ChunkedOutput$1.call(ChunkedOutput.java:192)
at org.glassfish.jersey.internal.Errors.process(Errors.java:315)
at org.glassfish.jersey.internal.Errors.process(Errors.java:242)
at org.glassfish.jersey.process.internal.RequestScope.runInScope(RequestScope.java:345)
at org.glassfish.jersey.server.ChunkedOutput.flushQueue(ChunkedOutput.java:192)
at org.glassfish.jersey.server.ChunkedOutput.write(ChunkedOutput.java:182)
at com.bpc.services.service.ObserverThread.run(MarketObserverThread.java:32)
at java.lang.Thread.run(Thread.java:745)
<<<<<<<SSE CLIENT HAS BEEN DEREGISTERED!
I have attempted to test an sse broadcaster as well. In this case I am not seeing any exceptions thrown, but the connection is closed once the first message has been received, leading me to believe it is something in the servlet forcing the connection to close. Can anyone advise me on how to debug this on the server-side?
I had a similar issue from what seems to be a long standing bug in Jersey's #Context injection for ExecutorService instances. In their current implementation of Sse (version 2.27),
class JerseySse implements Sse {
#Context
private ExecutorService executorService;
#Override
public OutboundSseEvent.Builder newEventBuilder() {
return new OutboundEvent.Builder();
}
#Override
public SseBroadcaster newBroadcaster() {
return new JerseySseBroadcaster(executorService);
}
}
the executorService field is never initialized, so the JerseySseBroadcaster raises a NullPointerException in my case. I worked around the bug by explicitly triggering the injection.
If you're using HK2 for CDI (Jersey's default), a rough sketch of a solution to the question above could look similar to the following:
#Singleton
#Path("...")
public class JmsPublisher {
private Sse sse;
private SseBroadcaster broadcaster;
private final ExecutorService executor;
private final BlockingQueue<String> jmsMessageQueue;
...
#Context
public void setSse(Sse sse, ServiceLocator locator) {
locator.inject(sse); // Inject sse.executorService
this.sse = sse;
this.broadcaster = sse.newBroadcaster();
}
...
#GET
#Path("/stream_data")
#Produces(MediaType.SERVER_SENT_EVENTS)
public void register(SseEventSink eventSink) {
broadcaster.register(eventSink);
}
...
#PostConstruct
private void postConstruct() {
executor.submit(() -> {
try {
while(true) {
String message = jmsMessageQueue.take();
broadcaster.broadcast(sse.newEventBuilder()
.mediaType(MediaType.APPLICATION_JSON_TYPE)
.data(String.class, message)
.build());
}
} catch(InterruptedException e) {
Thread.currentThread().interrupt();
}
});
}
#PreDestroy
private void preDestroy() {
executor.shutdownNow();
}
}

New log and lock files created when redeploying app

I'm logging in a JSF and ejb enviroment and every time I redeploy my app, the .lck files don't get deleted and a new log (and lock) file is created (logfilename.log.2, logfilename.log.3 ...)
This is how I get the log and add a filehandler in a JSF managed bean:
static {
// (...)
logger = Logger.getLogger("registrations");
FileHandler fh;
try {
// (...)
fh = new FileHandler(registerLogPath, true);
fh.setFormatter(new Formatter() {
#Override
public String format(LogRecord record) {
return ...
}
});
logger.addHandler(fh);
} catch (IOException | SecurityException ex) {
logger.log(Level.SEVERE, null, ex);
}
}
I tried to remove the handlers in the contextDestroyed method of ServletContextListener, but it seems that it is too late since Logger.getLogger("registrations").getHandlers() returns an empty array, so nothing to remove.
Can you help me how to make the logger use the file it used before redeploying?
I finally managed to solve this. Removing a handler is not enough, you have to call close() on the handler itself. Also, logger.getHandlers() works in a contextDestroyed method:
for (Handler h : logger.getHandlers()) {
h.close();
}

Detecting when a handler couldn't be started when embedding Jetty

I'm embedding Jetty in a similar manner as described here. When the RequestLogHandler can't open the specified logfile, it throws an exception which is unfortunately caught by org.eclipse.jetty.server.Server and swallowed (but logged first, at least). This means that there's no obvious way for me to tell if the log handler was started correctly.
Is there a way that I'm missing to detect when a handler couldn't start?
This idea is based on the implementation of WebAppContext where you can use WebAppContext.getUnavailableException() to determine whether the context was initialized successfully.
Simply replace the default implementation of Server and Context with your own:
public static class MyContext extends Context {
private Exception _exception;
#Override
protected void doStart() throws Exception {
try {
super.doStart();
} catch (final Exception e) {
_exception = e;
}
}
#Override
protected void doStop() throws Exception {
try {
super.doStop();
} finally {
_exception = null;
}
}
public Exception getException() {
return _exception;
}
}
public static class MyServer extends Server implements InitializingBean {
public void afterPropertiesSet() throws Exception {
start();
for (final Handler h : getHandlers()) {
if (h instanceof MyContext) {
final MyContext c = (MyContext) h;
if (c.getException() != null) {
throw new RuntimeException("failed to init context " + c.getDisplayName(),
c.getException());
}
}
}
}
}
In your beans.xml, simply replace org.mortbay.jetty.Server (and remove init-method="start") and org.mortbay.jetty.servlet.Context with your own implementations.
This code is for Jetty 6 though (as is the example you linked to), as that's what I have around. I didn't test it though, but it's pretty much the same as we are successfully using in conjunction with WebAppContext. In order to extend this to RequestLogHandler, you could either do the same for just any handler you are using or create a decorator to wrap any handler. You may want to look at org.mortbay.jetty.handler.HandlerWrapper for this purpose.
How about modifying the jetty code? You could add some simple println statements in strategic places in the RequestLogHandler which would indicate to you whether or not the handler was started.

Categories

Resources