I'm trying to unit test an advice on the poller which blocks execution of the mongo channel adapter until a certain condition is met (=all messages from this batch are processed).
The flow looks as follow:
IntegrationFlows.from(MongoDb.reactiveInboundChannelAdapter(mongoDbFactory,
new Query().with(Sort.by(Sort.Direction.DESC, "modifiedDate")).limit(1))
.collectionName("metadata")
.entityClass(Metadata.class)
.expectSingleResult(true),
e -> e.poller(Pollers.fixedDelay(Duration.ofSeconds(pollingIntervalSeconds))
.advice(this.advices.waitUntilCompletedAdvice())))
.handle((p, h) -> {
this.advices.waitUntilCompletedAdvice().setWait(true);
return p;
})
.handle(doSomething())
.channel(Channels.DOCUMENT_HEADER.name())
.get();
And the following advice bean:
#Bean
public WaitUntilCompletedAdvice waitUntilCompletedAdvice() {
DynamicPeriodicTrigger trigger = new DynamicPeriodicTrigger(Duration.ofSeconds(1));
return new WaitUntilCompletedAdvice(trigger);
}
And the advice itself:
public class WaitUntilCompletedAdvice extends SimpleActiveIdleMessageSourceAdvice {
AtomicBoolean wait = new AtomicBoolean(false);
public WaitUntilCompletedAdvice(DynamicPeriodicTrigger trigger) {
super(trigger);
}
#Override
public boolean beforeReceive(MessageSource<?> source) {
if (getWait())
return false;
return true;
}
public boolean getWait() {
return wait.get();
}
public void setWait(boolean newWait) {
if (getWait() == newWait)
return;
while (true) {
if (wait.compareAndSet(!newWait, newWait)) {
return;
}
}
}
}
I'm using the following test for testing the flow:
#Test
public void testClaimPoollingAdapterFlow() throws Exception {
// given
ArgumentCaptor<Message<?>> captor = messageArgumentCaptor();
CountDownLatch receiveLatch = new CountDownLatch(1);
MessageHandler mockMessageHandler = mockMessageHandler(captor).handleNext(m -> receiveLatch.countDown());
this.mockIntegrationContext.substituteMessageHandlerFor("retrieveDocumentHeader", mockMessageHandler);
LocalDateTime modifiedDate = LocalDateTime.now();
ProcessingMetadata data = Metadata.builder()
.modifiedDate(modifiedDate)
.build();
assert !this.advices.waitUntilCompletedAdvice().getWait();
// when
itf.getInputChannel().send(new GenericMessage<>(Mono.just(data)));
// then
assertThat(receiveLatch.await(10, TimeUnit.SECONDS)).isTrue();
verify(mockMessageHandler).handleMessage(any());
assertThat(captor.getValue().getPayload()).isEqualTo(modifiedDate);
assert this.advices.waitUntilCompletedAdvice().getWait();
}
Which works fine but when I send another message to the input channel, it still processes the message without respecting the advice.
Is it intended behaviour? If so, how can I verify using unit test that the poller is really waiting for this advice?
itf.getInputChannel().send(new GenericMessage<>(Mono.just(data)));
That bypasses the poller and sends the message directly.
You can unit test the advice has been configured by calling beforeReceive() from your test
Or you can create a dummy test flow with the same advice
IntegationFlows.from(() -> "foo", e -> e.poller(...))
...
And verify that just one message is sent.
Example implementation:
#Test
public void testWaitingActivate() {
// given
this.advices.waitUntilCompletedAdvice().setWait(true);
// when
Message<ProcessingMetadata> receive = (Message<ProcessingMetadata>) testChannel.receive(3000);
// then
assertThat(receive).isNull();
}
#Test
public void testWaitingInactive() {
// given
this.advices.waitUntilCompletedAdvice().setWait(false);
// when
Message<ProcessingMetadata> receive = (Message<ProcessingMetadata>) testChannel.receive(3000);
// then
assertThat(receive).isNotNull();
}
#Configuration
#EnableIntegration
public static class Config {
#Autowired
private Advices advices;
#Bean
public PollableChannel testChannel() {
return new QueueChannel();
}
#Bean
public IntegrationFlow fakeFlow() {
this.advices.waitUntilCompletedAdvice().setWait(true);
return IntegrationFlows.from(() -> "foo", e -> e.poller(Pollers.fixedDelay(Duration.ofSeconds(1))
.advice(this.advices.waitUntilCompletedAdvice()))).channel("testChannel").get();
}
}
Related
I want to build an API based on Futures (from java.util.concurrent) that is powered by a custom protocol on top of Netty (version 4). Basic idea is to write a simple library that would abstract the underlying Netty implementation and make it easier to make requests.
Using this library, one should be able to write something like this:
Request req = new Request(...);
Future<Response> responseFuture = new ServerIFace(host, port).call(req);
// For example, let's block until this future is resolved
Reponse res = responseFuture.get().getResult();
Underneath this code, a Netty client is connected
public class ServerIFace {
private Bootstrap bootstrap;
private EventLoopGroup workerGroup;
private String host;
private int port;
public ServerIFace(String host, int port) {
this.host = host;
this.port = port;
this.workerGroup = new NioEventLoopGroup();
bootstrap();
}
private void bootstrap() {
bootstrap = new Bootstrap();
bootstrap.group(workerGroup);
bootstrap.channel(NioSocketChannel.class);
bootstrap.handler(new ChannelInitializer<SocketChannel>() {
#Override
protected void initChannel(SocketChannel ch) throws Exception {
ch.pipeline().addLast(new ObjectEncoder());
ch.pipeline().addLast(new ObjectDecoder(ClassResolvers.cacheDisabled(Response.class.getClassLoader())));
ch.pipeline().addLast("response", new ResponseReceiverChannelHandler());
}
});
}
public Future<Response> call(final Request request) throws InterruptedException {
CompletableFuture<Response> responseFuture = new CompletableFuture<>();
Channel ch = bootstrap.connect(host, port).sync().channel();
ch.writeAndFlush(request).addListener((f) -> {
if (f.isSuccess()) {
System.out.println("Wrote successfully");
} else {
f.cause().printStackTrace();
}
});
ChannelFuture closeFuture = ch.closeFuture();
// Have to 'convert' ChannelFuture to java.util.concurrent.Future
closeFuture.addListener((f) -> {
if (f.isSuccess()) {
// How to get this response?
Response response = ((ResponseReceiverChannelHandler) ch.pipeline().get("response")).getResponse();
responseFuture.complete(response);
} else {
f.cause().printStackTrace();
responseFuture.cancel(true);
}
ch.close();
}).sync();
return responseFuture;
}
}
Now, as you can see, in order to abstract Netty's inner ChannelFuture, I have to 'convert' it to Java's Future (I'm aware that ChannelFuture is derived from Future, but that information doesn't seem useful at this point).
Right now, I'm capturing this Response object in the last handler of my inbound part of the client pipeline, the ResponseReceiverChannelHandler.
public class ResponseReceiverChannelHandler extends ChannelInboundHandlerAdapter {
private Response response = null;
#Override
public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
this.response = (Response)msg;
ctx.close();
}
public Response getResponse() {
return response;
}
}
Since I'm new to Netty and these things in general, I'm looking for a cleaner, thread-safe way of delivering this object to the API user.
Correct me if I'm wrong, but none of the Netty examples show how to achieve this, and most of the Client examples just print out whatever they get from Server.
Please note that my main goal here is to learn more about Netty, and that this code has no production purposes.
For the reference (although I don't think it's that relevant) here's the Server code.
public class Server {
public static class RequestProcessorHandler extends ChannelInboundHandlerAdapter {
#Override
public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
ChannelFuture future;
if (msg instanceof Request) {
Request req = (Request)msg;
Response res = some function of req
future = ctx.writeAndFlush(res);
} else {
future = ctx.writeAndFlush("Error, not a request!");
}
future.addListener((f) -> {
if (f.isSuccess()) {
System.out.println("Response sent!");
} else {
System.out.println("Response not sent!");
f.cause().printStackTrace();
}
});
}
}
public int port;
public Server(int port) {
this.port = port;
}
public void run() throws Exception {
EventLoopGroup bossGroup = new NioEventLoopGroup();
EventLoopGroup workerGroup = new NioEventLoopGroup();
try {
ServerBootstrap b = new ServerBootstrap();
b.group(bossGroup, workerGroup)
.channel(NioServerSocketChannel.class)
.childHandler(new ChannelInitializer<SocketChannel>() {
#Override
public void initChannel(SocketChannel ch) throws Exception {
ch.pipeline().addLast(new ObjectDecoder(ClassResolvers.cacheDisabled(Request.class.getClassLoader())));
ch.pipeline().addLast(new ObjectEncoder());
// Not really shutting down this threadpool but it's ok for now
ch.pipeline().addLast(new DefaultEventExecutorGroup(2), new RequestProcessorHandler());
}
})
.option(ChannelOption.SO_BACKLOG, 128)
.childOption(ChannelOption.SO_KEEPALIVE, true);
ChannelFuture f = b.bind(port).sync();
f.channel().closeFuture().sync();
} finally {
workerGroup.shutdownGracefully();
bossGroup.shutdownGracefully();
}
}
public static void main(String[] args) throws Exception {
int port;
if (args.length > 0) {
port = Integer.parseInt(args[0]);
} else {
port = 8080;
}
new Server(port).run();
}
}
[long description warning]
I'm running some cucumber tests which have to be executed intercalated a defined server - for instance:
a.feature -> JBoss Server 1 | b.feature -> JBoss Serv. 2 | c.feature -> JB1 | etc.
For that, I created a hypothetical ExecutorService like this:
final ExecutorService executorService = Executors.newFixedThreadPool(2); //numberOfServers
for (Runnable task : tasks) {
executorService.execute(task);
}
executorService.shutdown();
try {
executorService.awaitTermination(1000, TimeUnit.SECONDS);
} catch (InterruptedException e) {
//doX();
}
The way that I manage about how will be the server chosen as liable to execute is:
inside of my Runnable class created for the executorService, I pass as a parameter a instanceId to a TestNG (XmlTest class) as below:
#Override
public void run() {
setupTest().run();
}
private TestNG setupTest() {
TestNG testNG = new TestNG();
XmlSuite xmlSuite = new XmlSuite();
XmlTest xmlTest = new XmlTest(xmlSuite);
xmlTest.setName(//irrelevant);
xmlTest.addParameter("instanceId", String.valueOf(instanceId));
xmlTest.setXmlClasses(..........);
testNG.setXmlSuites(..........);
return testNG;
}
Then, I get this just fine in a class that extends TestNgCucumberAdaptor:
#BeforeTest
#Parameters({"instanceId"})
public void setInstanceId(#Optional("") String instanceId) {
if (!StringUtils.isEmpty(instanceId)) {
this.instanceId = Integer.valueOf(instanceId);
}
}
And inside a #BeforeClass I'm populating a Pojo with this instanceId and setting the Pojo in a threadLocal attribute of another class. So far, so good.
public class CurrentPojoContext {
private static final ThreadLocal<PojoContext> TEST_CONTEXT = new ThreadLocal<PojoContext>();
...
public static PojoContext getContext(){
TEST_CONTEXT.get();
}
Now the problem really starts - I'm using Guice (Cucumber guice as well) in a 3rd class, injecting this pojo object that contains the instanceId. The example follows:
public class Environment {
protected final PojoContext pojoContext;
#Inject
public Environment() {
this.pojoContext = CurrentPojoContext.getContext();
}
public void foo() {
print(pojoContext.instanceId); // output: 1
Another.doSomething(pojoContext);
}
class Another{
public String doSomething(PojoContext p){
print(p.instanceId); // output: 2
}
}
}
Here it is not every time like this the outputs (1 and 2) but from time to time, I realized that the execution of different threads is messing with the attribute pojoContext. I know that is a little confusing, but my guess is that the Guice Injector is not thread-safe for this scenario - it might be a long shot, but I'd appreciate if someone else takes a guess.
Regards
Well, just in order to provide a solution for someone else, my solution was the following:
Create a class that maintains a Map with an identifier (unique and thread-safe one) as the key and a Guice Injector as value;
Inside my instantiation of Guice injector, I created my own module
Guice.createInjector(Stage.PRODUCTION, MyOwnModules.SCENARIO, new RandomModule());
and for this module:
public class MyOwnModules {
public static final Module SCENARIO = new ScenarioModule(MyOwnCucumberScopes.SCENARIO);
}
the scope defined here provides the following:
public class MyOwnCucumberScopes {
public static final ScenarioScope SCENARIO = new ParallelScenarioScope();
}
To sum up, the thread-safe will be in the ParallelScenarioScope:
public class ParallelScenarioScope implements ScenarioScope {
private static final Logger LOGGER = Logger.getLogger(ParallelScenarioScope.class);
private final ThreadLocal<Map<Key<?>, Object>> threadLocalMap = new ThreadLocal<Map<Key<?>, Object>>();
#Override
public <T> Provider<T> scope(final Key<T> key, final Provider<T> unscoped) {
return new Provider<T>() {
public T get() {
Map<Key<?>, Object> scopedObjects = getScopedObjectMap(key);
#SuppressWarnings("unchecked")
T current = (T) scopedObjects.get(key);
if (current == null && !scopedObjects.containsKey(key)) {
current = unscoped.get();
scopedObjects.put(key, current);
}
return current;
}
};
}
protected <T> Map<Key<?>, Object> getScopedObjectMap(Key<T> key) {
Map<Key<?>, Object> map = threadLocalMap.get();
if (map == null) {
throw new OutOfScopeException("Cannot access " + key + " outside of a scoping block");
}
return map;
}
#Override
public void enterScope() {
checkState(threadLocalMap.get() == null, "A scoping block is already in progress");
threadLocalMap.set(new ConcurrentHashMap<Key<?>, Object>());
}
#Override
public void exitScope() {
checkState(threadLocalMap.get() != null, "No scoping block in progress");
threadLocalMap.remove();
}
private void checkState(boolean expression, String errorMessage) {
if (!expression) {
LOGGER.info("M=checkState, Will throw exception: " + errorMessage);
throw new IllegalStateException(errorMessage);
}
}
}
Now the gotcha is just to be careful regarding the #ScenarioScoped and the code will work as expected.
I have a Spring integration application that normally polls daily for a file via SFTP using a cron trigger. But if it doesn't find the file it expects, it should poll every x minutes via a periodic trigger until y attempts. To do this I use the following component:
#Component
public class RetryCompoundTriggerAdvice extends AbstractMessageSourceAdvice {
private final static Logger logger = LoggerFactory.getLogger(RetryCompoundTriggerAdvice.class);
private final CompoundTrigger compoundTrigger;
private final Trigger override;
private final ApplicationProperties applicationProperties;
private final Mail mail;
private int attempts = 0;
public RetryCompoundTriggerAdvice(CompoundTrigger compoundTrigger,
#Qualifier("secondaryTrigger") Trigger override,
ApplicationProperties applicationProperties,
Mail mail) {
this.compoundTrigger = compoundTrigger;
this.override = override;
this.applicationProperties = applicationProperties;
this.mail = mail;
}
#Override
public boolean beforeReceive(MessageSource<?> source) {
return true;
}
#Override
public Message<?> afterReceive(Message<?> result, MessageSource<?> source) {
final int maxOverrideAttempts = applicationProperties.getMaxFileRetry();
attempts++;
if (result == null && attempts < maxOverrideAttempts) {
logger.info("Unable to find load file after " + attempts + " attempt(s). Will reattempt");
this.compoundTrigger.setOverride(this.override);
} else if (result == null && attempts >= maxOverrideAttempts) {
mail.sendAdminsEmail("Missing File");
attempts = 0;
this.compoundTrigger.setOverride(null);
}
else {
attempts = 0;
this.compoundTrigger.setOverride(null);
logger.info("Found load file");
}
return result;
}
public void setOverrideTrigger() {
this.compoundTrigger.setOverride(this.override);
}
public CompoundTrigger getCompoundTrigger() {
return compoundTrigger;
}
}
If a file doesn't exist, this works great. That is, the override (i.e. periodic trigger) takes effect and polls every x minutes until y attempts.
However, if a file does exist but it's not the expected file (e.g. the data is at the wrong date), another class (that reads the file) calls the setOverrideTrigger of the RetryCompoundTriggerAdvice class. But afterReceive is not subsequently called at every x minutes. Why would this be?
Here's more of the application code:
SftpInboundFileSynchronizer:
#Bean
public SftpInboundFileSynchronizer sftpInboundFileSynchronizer() {
SftpInboundFileSynchronizer fileSynchronizer = new SftpInboundFileSynchronizer(sftpSessionFactory());
fileSynchronizer.setDeleteRemoteFiles(false);
fileSynchronizer.setRemoteDirectory(applicationProperties.getSftpDirectory());
CompositeFileListFilter<ChannelSftp.LsEntry> compositeFileListFilter = new CompositeFileListFilter<ChannelSftp.LsEntry>();
compositeFileListFilter.addFilter(new SftpPersistentAcceptOnceFileListFilter(store, "sftp"));
compositeFileListFilter.addFilter(new SftpSimplePatternFileListFilter(applicationProperties.getLoadFileNamePattern()));
fileSynchronizer.setFilter(compositeFileListFilter);
fileSynchronizer.setPreserveTimestamp(true);
return fileSynchronizer;
}
Session factory is:
#Bean
public SessionFactory<LsEntry> sftpSessionFactory() {
DefaultSftpSessionFactory sftpSessionFactory = new DefaultSftpSessionFactory();
sftpSessionFactory.setHost(applicationProperties.getSftpHost());
sftpSessionFactory.setPort(applicationProperties.getSftpPort());
sftpSessionFactory.setUser(applicationProperties.getSftpUser());
sftpSessionFactory.setPassword(applicationProperties.getSftpPassword());
sftpSessionFactory.setAllowUnknownKeys(true);
return new CachingSessionFactory<LsEntry>(sftpSessionFactory);
}
The SftpInboundFileSynchronizingMessageSource is set to poll using the compound trigger.
#Bean
#InboundChannelAdapter(autoStartup="true", channel = "sftpChannel", poller = #Poller("pollerMetadata"))
public SftpInboundFileSynchronizingMessageSource sftpMessageSource() {
SftpInboundFileSynchronizingMessageSource source =
new SftpInboundFileSynchronizingMessageSource(sftpInboundFileSynchronizer());
source.setLocalDirectory(applicationProperties.getScheduledLoadDirectory());
source.setAutoCreateLocalDirectory(true);
CompositeFileListFilter<File> compositeFileFilter = new CompositeFileListFilter<File>();
compositeFileFilter.addFilter(new LastModifiedFileListFilter());
compositeFileFilter.addFilter(new FileSystemPersistentAcceptOnceFileListFilter(store, "dailyfilesystem"));
source.setLocalFilter(compositeFileFilter);
source.setCountsEnabled(true);
return source;
}
#Bean
public PollerMetadata pollerMetadata(RetryCompoundTriggerAdvice retryCompoundTriggerAdvice) {
PollerMetadata pollerMetadata = new PollerMetadata();
List<Advice> adviceChain = new ArrayList<Advice>();
adviceChain.add(retryCompoundTriggerAdvice);
pollerMetadata.setAdviceChain(adviceChain);
pollerMetadata.setTrigger(compoundTrigger());
pollerMetadata.setMaxMessagesPerPoll(1);
return pollerMetadata;
}
#Bean
public CompoundTrigger compoundTrigger() {
CompoundTrigger compoundTrigger = new CompoundTrigger(primaryTrigger());
return compoundTrigger;
}
#Bean
public CronTrigger primaryTrigger() {
return new CronTrigger(applicationProperties.getSchedule());
}
#Bean
public PeriodicTrigger secondaryTrigger() {
return new PeriodicTrigger(applicationProperties.getRetryInterval());
}
Update
Here's the message handler:
#Bean
#ServiceActivator(inputChannel = "sftpChannel")
public MessageHandler dailyHandler(SimpleJobLauncher jobLauncher, Job job, Mail mail) {
JobRunner jobRunner = new JobRunner(jobLauncher, job, store, mail);
jobRunner.setDaily("true");
jobRunner.setOverwrite("false");
return jobRunner;
}
JobRunner kicks off a Spring Batch job. After processing the job, my application looks to see if the file had the data it expected for the day. If not, it is setting the override trigger.
That's the way triggers work - you only get an opportunity to change the trigger when the trigger fires.
Since you reset to the cron trigger, the next opportunity for change is when that trigger fires (if the poller thread is released by the downstream flow before changing the trigger).
Are you handing off the file to another thread (queue channel or executor)? If not, I would expect any changes to the trigger should be applied, because nextExecutionTime() will not be called until the downstream flow returns.
If there's a thread handoff, you have no opportunity to change the trigger.
I'm running an application that process tasks using spring integration.
I'd like to make it process multiple tasks concurrently but any attempt failed so far.
My configuration is:
ReactorConfiguration.java
#Configuration
#EnableAutoConfiguration
public class ReactorConfiguration {
#Bean
Environment reactorEnv() {
return new Environment();
}
#Bean
Reactor createReactor(Environment env) {
return Reactors.reactor()
.env(env)
.dispatcher(Environment.THREAD_POOL)
.get();
}
}
TaskProcessor.java
#MessagingGateway(reactorEnvironment = "reactorEnv")
public interface TaskProcessor {
#Gateway(requestChannel = "routeTaskByType", replyChannel = "")
Promise<Result> processTask(Task task);
}
IntegrationConfiguration.java (simplified)
#Bean
public IntegrationFlow routeFlow() {
return IntegrationFlows.from(MessageChannels.executor("routeTaskByType", Executors.newFixedThreadPool(10)))
.handle(Task.class, (payload, headers) -> {
logger.info("Task submitted!" + payload);
payload.setRunning(true);
//Try-catch
Thread.sleep(999999);
return payload;
})
.route(/*...*/)
.get();
}
My testing code can be simplified like this:
Task task1 = new Task();
Task task2 = new Task();
Promise<Result> resultPromise1 = taskProcessor.processTask(task1).flush();
Promise<Result> resultPromise2 = taskProcessor.processTask(task2).flush();
while( !task1.isRunning() || !task2.isRunning() ){
logger.info("Task2: {}, Task2: {}", task1, task2);
Thread.sleep(1000);
}
logger.info("Yes! your tasks are running in parallel!");
But unfortunately, the last log line, will never get executed!
Any ideas?
Thanks a lot
Well, I've reproduced it just with simple Reactor test-case:
#Test
public void testParallelPromises() throws InterruptedException {
Environment environment = new Environment();
final AtomicBoolean first = new AtomicBoolean(true);
for (int i = 0; i < 10; i++) {
final Promise<String> promise = Promises.task(environment, () -> {
if (!first.getAndSet(false)) {
try {
Thread.sleep(1000);
}
catch (InterruptedException e) {
e.printStackTrace();
}
}
return "foo";
}
);
String result = promise.await(10, TimeUnit.SECONDS);
System.out.println(result);
assertNotNull(result);
}
}
(It is with Reactor-2.0.6).
The problem is because of:
public static <T> Promise<T> task(Environment env, Supplier<T> supplier) {
return task(env, env.getDefaultDispatcher(), supplier);
}
where DefaultDispatcher is RingBufferDispatcher extends SingleThreadDispatcher.
Since the #MessagingGateway is based on the request/reply scenario, we are waiting for reply within that RingBufferDispatcher's Thread. Since you don't return reply there (Thread.sleep(999999);), we aren't able to accept the next event within RingBuffer.
Your dispatcher(Environment.THREAD_POOL) doesn't help here because it doesn't affect the Environment. You should consider to use reactor.dispatchers.default = threadPoolExecutor property. Something like this file: https://github.com/reactor/reactor/blob/2.0.x/reactor-net/src/test/resources/META-INF/reactor/reactor-environment.properties#L46.
And yes: upgrade, please, to the latest Reactor.
I have a springframework application in which I would like to add a transaction listener to a transaction which is currently in progress. The motivation is to trigger a post commit action which notifies downstream systems. I am using #Transactional to wrap a transaction around some service method -- which is where I want to create/register the post transaction listener. I want to do something "like" the following.
public class MyService {
#Transaction
public void doIt() {
modifyObjects();
// something like this
getTransactionManager().registerPostCommitAction(new
TransactionSynchronizationAdapter() {
public void afterCommit() {
notifyDownstream();
}
});
}
}
Spring has a TransactionSynchronization interface and adapter class which seems exactly what I want; however it is not immediately clear how to register one dynamically with either the current transaction, or the transaction manager. I would rather not subclass JtaTransactionManager if I can avoid it.
Q: Has anyone done this before.
Q: what is the simplest way to register my adapter?
Actually it was not as hard as I thought; spring has a static helper class that puts the 'right' stuff into the thread context.
TransactionSynchronizationManager.registerSynchronization(
new TransactionSynchronizationAdapter() {
#Override
public void afterCommit() {
s_logger.info("TRANSACTION COMPLETE!!!");
}
}
);
you could use an aspect to match transactional methods aspect in your service to accomplish this:
#Aspect
public class AfterReturningExample {
#AfterReturning("execution(* com.mypackage.MyService.*(..))")
public void afterReturning() {
// ...
}
}
Here is a more complete solution I did for a similar problem that with wanting my messages sent after transactions are committed (I could have used RabbitMQ TX but they are rather slow).
public class MessageBusUtils {
public static Optional<MessageBusResourceHolder> getTransactionalResourceHolder(TxMessageBus messageBus) {
if ( ! TransactionSynchronizationManager.isActualTransactionActive()) {
return Optional.absent();
}
MessageBusResourceHolder o = (MessageBusResourceHolder) TransactionSynchronizationManager.getResource(messageBus);
if (o != null) return Optional.of(o);
o = new MessageBusResourceHolder();
TransactionSynchronizationManager.bindResource(messageBus, o);
o.setSynchronizedWithTransaction(true);
if (TransactionSynchronizationManager.isSynchronizationActive()) {
TransactionSynchronizationManager.registerSynchronization(new MessageBusResourceSynchronization(o, messageBus));
}
return Optional.of(o);
}
private static class MessageBusResourceSynchronization extends ResourceHolderSynchronization<MessageBusResourceHolder, TxMessageBus> {
private final TxMessageBus messageBus;
private final MessageBusResourceHolder holder;
public MessageBusResourceSynchronization(MessageBusResourceHolder resourceHolder, TxMessageBus resourceKey) {
super(resourceHolder, resourceKey);
this.messageBus = resourceKey;
this.holder = resourceHolder;
}
#Override
protected void cleanupResource(MessageBusResourceHolder resourceHolder, TxMessageBus resourceKey,
boolean committed) {
resourceHolder.getPendingMessages().clear();
}
#Override
public void afterCompletion(int status) {
if (status == TransactionSynchronization.STATUS_COMMITTED) {
for (Object o : holder.getPendingMessages()) {
messageBus.post(o, false);
}
}
else {
holder.getPendingMessages().clear();
}
super.afterCompletion(status);
}
}
}
public class MessageBusResourceHolder extends ResourceHolderSupport {
private List<Object> pendingMessages = Lists.newArrayList();
public void addMessage(Object message) {
pendingMessages.add(message);
}
protected List<Object> getPendingMessages() {
return pendingMessages;
}
}
Now in your class where you actually send the message you will do
#Override
public void postAfterCommit(Object o) {
Optional<MessageBusResourceHolder> holder = MessageBusTxUtils.getTransactionalResourceHolder(this);
if (holder.isPresent()) {
holder.get().addMessage(o);
}
else {
post(o, false);
}
}
Sorry for the long winded coding samples but hopefully that will show someone how to do something after a commit.
Does it make sense to override the transaction manager on the commit and rollback methods, calling super.commit() right at the beginning.