We Keep Coding

SimpleAsyncTaskExecutor with a queue

I have a need to run some threads concurrently, but need to force each process to run in a new Thread (this is due to some ThreadLocal bleed that I don't have full control over). To do so, I have been using the SimpleAsyncTaskExecutor. However, the issue with this is that it doesn't maintain a queue that allows new tasks to be submitted once it's reached the concurrency limit. What I really need to do is have functionality like the SimpleAsyncTaskExecutor but where tasks can still be submitted even after the concurrency limit has been reached - I just want those tasks to wait in the queue until another slot frees up. This is what I have right now:
SimpleAsyncTaskExecutor taskExecutor = new SimpleAsyncTaskExecutor();
taskExecutor.setConcurrencyLimit(maxThreads);
return taskExecutor;
Is there some out-of-the-box solution for this, or do I need to write something custom?

To ensure you need to execute every task in a new Thread, You are basically against use of any ThreadPool (ThreadLocal behavior in a ThreadPool is something you need to get rid of, sooner or later).
To overcome this, you can simply produce something like this,
class ThreadPerTaskExecutor implements Executor {
public void execute(Runnable r) {
Thread t = new Thread(r);
t.start();
try {
t.join();
} catch (InterruptedException e) {
e.printStackTrace();
}
}
}
which executes the Runnable always in a new Thread.
Coming to a crude implementation, We can do something like
final Executor executor = new ThreadPerTaskExecutor();
final ExecutorService service = Executors.newFixedThreadPool(3);
for (int i = 0; i < 100; i++) {
service.submit(new Runnable() {
public void run() {
try {
System.out.println("Executed inside Thread pool with concurrency level 3"
+ Thread.currentThread().toString());
executor.execute(new Runnable() {
public void run() {
try {
Thread.sleep(3000); //Some expensive operations here.
System.out.println(
"Executed inside new Thread always" + Thread.currentThread().toString());
} catch (InterruptedException e) {
e.printStackTrace();
}
}
});
} catch (Exception e) {
e.printStackTrace();
}
}
});
}
This can be improved with lambdas as well after Java 8. Hope this sheds the basic idea.

Is there some out-of-the-box solution for this, or do I need to write something custom?
I think there is no out-of-the-box solution for this, and you need to write your own code for this.
You can extend the SimpleAsyncTaskExecutor for simpler/quicker implementation. Example:
public class SimpleAsyncQueueTaskExecutor extends SimpleAsyncTaskExecutor {
private Queue<Runnable> queue = new ConcurrentLinkedQueue<Runnable>();
private AtomicInteger concurrencyValue = new AtomicInteger(0);
private void checkAndExecuteFromQueue() {
int count = concurrencyValue.get();
if (isThrottleActive() && !queue.isEmpty() &&
(count < getConcurrencyLimit())) {
Runnable task = queue.poll();
concurrencyValue.incrementAndGet();
doExecute(new ConcurrencyThrottlingRunnable(task));
}
}
private void afterExecute(Runnable task) {
queue.remove(task);
concurrencyValue.decrementAndGet();
// Check and execute other tasks
checkAndExecuteFromQueue();
}
#Override
public void execute(Runnable task, long startTimeout) {
Assert.notNull(task, "Runnable must not be null");
if (isThrottleActive() && startTimeout > TIMEOUT_IMMEDIATE) {
queue.offer(task);
checkAndExecuteFromQueue();
} else {
doExecute(task);
}
}
private class ConcurrencyThrottlingRunnable implements Runnable {
private final Runnable target;
public ConcurrencyThrottlingRunnable(Runnable target) {
this.target = target;
}
#Override
public void run() {
try {
this.target.run();
}
finally {
afterExecute(this.target);
}
}
}
This example code just add a queue, and override the execute method.
Hope this help.

Java Thread with persistent resources

I have a pool of worker threads (an ExecutorService).
This pool is used to run shell commands.
I use a shell (/bin/sh) rather than creating a process for the executable directly, because I use shell redirects (>) to write the output directly to disk, without having to pass through the JVM, as well as some other niceties.
Spawning a shell process takes 2-3 milliseconds.
I want each thread to keep a shell process to avoid the overhead of starting it.
How do I allow each thread to own a process?
I am thinking of using a ThreadFactory with thread locals.
class ThreadFactory {
Thread newThread(Runnable r) {
return new Thread(new Runnable() {
Process process = Runtime.getRuntime().exec("/bin/sh")
try {
// store process as thread local here
r.run(); // then r can access thread local
} catch(Exception e) {
try {
process.close();
} catch(Exception e) {
}
throw e;
}
});
}
}
(Alternatively, I could subclass Thread and cast Thread.currentThread() to that class in my Runnable.)
Is this a good approach to solving this problem?

I would keep the Process reference in a ProcessRunnable that continuously executes commands. I think is more clear than using a ThreadLocal and a ThreadFactory. Something like this:
public class ShellCommandExecutor {
private int concurrency = 10;
private int capacity = 100;
private ExecutorService service = Executors.newFixedThreadPool(concurrency);
private BlockingQueue<String> commandsQueue = new LinkedBlockingQueue<>(capacity);
public void start() {
for (int i = 0; i < concurrency; i++)
service.submit(new Runnable() {
#Override
public void run() {
//todo deal with ioexception
Process process = Runtime.getRuntime().exec("/bin/sh");
while (!Thread.currentThread().isInterrupted()) {
try {
String command = commandsQueue.take();
//todo execute commands using the same process per thread
}
catch (InterruptedException e) {
Thread.currentThread().interrupt();
}
}
}
});
}
public void executeCommand(String command) throws InterruptedException {
commandsQueue.put(command);
}
public void shutdown() {
service.shutdownNow();
}
}
EDIT: a solution with thread local that should work easily with cached thread pools:
public class ShellCommandExecutor2 {
//todo limit queue
private ExecutorService service = Executors.newCachedThreadPool();
public void executeCommand(final String command) throws InterruptedException {
service.submit(new Runnable() {
#Override
public void run() {
Process process = ThreadLocalProcessFactory.get();
//todo execute command
}
});
}
public void shutdown() {
service.shutdownNow();
}
private static class ThreadLocalProcessFactory {
private static final ThreadLocal<Process> processThreadLocal =
new ThreadLocal<Process>() {
#Override protected Process initialValue() {
try {
return Runtime.getRuntime().exec("/bin/sh");
}
catch (IOException e) {
e.printStackTrace();
return null;
}
}
};
static Process get() {
return processThreadLocal.get();
}
}
}

Lightweight long-running method cancel pattern for Java

Is there a lightweight pattern to cancel long running method, which replaces code like this:
public void longComputations() {
... first step...
if ( cancelled ) {
rollbackWork();
return;
}
... second step...
if ( cancelled ) {
rollbackWork();
return;
}
... third step...
if ( cancelled ) {
rollbackWork();
return;
}
}
I know that I can make a Task class, subdivide steps to task objects, make a queue and just do tasks step by steps in loop with cancelation checking, but I'm just wondering is there any simple code-structure pattern for such situation.

I am not aware of such a mechanism. Since you have to track your work in order to be able to perform rollbackWork(), a well-designed object-oriented solution is your best choice anyway, if you want to further evolve this logic! Typically, such a scenario could be implemented using the command pattern, which I still find pretty lightweight:
// Task or Command
public interface Command {
void redo();
void undo();
}
A scheduler or queue could then take care of executing such task / command implementations, and of rolling them back in order.

How about this edit, not a pattern though? Exceptions are very cheap, so it should be fast.
public void caller(){
try{
longComputations();
} catch (MeaningfulRuntimeException e){
rollbackWork(e.getStep());
}
}
public void longComputations() {
... first step...
checkStatus(1);
... second step...
checkStatus(2);
... third step...
checkStatus(3);
}
public void checkStatus(int step){
if ( cancelled ) {
... we may rollback here or throw an exception ...
throw MeaningfulRuntimeException(step);
}
}

If the steps call methods which throw InterruptedException then you can use Thread.interrupt(). You will still need to maintain enough state information to do the rollback properly.
If the steps cannot be interrupted this way, you should not consider relying on the deprecated Thread.stop() mechanism since it is inherently unsafe.
It seems that either way it makes sense to do exactly what you described: encapsulate this workflow logic in a class independent of the computation steps. It should support cancellation and/or interruption and accept a bunch of tasks to be executed. Note that the tasks to be fed into the workflow should provide at least two methods: one to perform the computation and one to roll it back.

You might want to consider using the java.util.concurrent package. You need to wrap your working steps as Callables (or Runnables).
public class InterruptibleTest {
public static void main(String[] args) { try {
final ExecutorService queue = Executors.newFixedThreadPool(1);
queue.submit(new Callable<Void>() { #Override public Void call() { busyWait(1000); return null; } });
queue.submit(new Callable<Void>() { #Override public Void call() { busyWait(1000); return null; } });
queue.submit(new Callable<Void>() { #Override public Void call() { busyWait(1000); return null; } });
final AtomicBoolean cancelled = new AtomicBoolean();
new Thread() { #Override public void run() {
try { Thread.sleep(1500); } catch (InterruptedException ex) { }
queue.shutdownNow();
cancelled.set(true);
}
}.run();
if (cancelled.get()) { rollback(); }
queue.shutdown();
System.out.println("Finished");
} catch (Exception ex) { ex.printStackTrace(System.err); } }
public synchronized static void busyWait(int millis) {
System.out.println("Start");
long until = System.currentTimeMillis() + millis;
while (System.currentTimeMillis() < until) { }
System.out.println("Stopped");
}
public synchronized static void rollback() {
System.out.println("Rollback!");
}
}
Note that shutdownNow() might call interrupt() on the currently executing work thread. You will probably also need to synchronize your rollback() because shutdownNow() returns before non-interruptible code finishes execution.

How to notify another thread

I want to know the best way how to notify another thread. For example, I have a background thread:
public void StartBackgroundThread(){
new Thread(new Runnable() {
#Override
public void run() {
//Do something big...
//THEN HOW TO NOTIFY MAIN THREAD?
}
}).start();
}
When it finished it has to notify main thread? If somebody knows the best way how to do this I'll appreciate it!

The typical answer is a BlockingQueue. Both BackgroundThread (often called the Producer) and MainThread (often called the Consumer) share a single instance of the queue (perhaps they get it when they are instantiated). BackgroundThread calls queue.put(message) each time it has a new message and MainThread calls 'queue.take()which will block until there's a message to receive. You can get fancy with timeouts and peeking but typically people want aBlockingQueueinstance such asArrayBlockingQueue`.

Purely based on your question you could do this:
public class test
{
Object syncObj = new Object();
public static void main(String args[])
{
new test();
}
public test()
{
startBackgroundThread();
System.out.println("Main thread waiting...");
try
{
synchronized(syncObj)
{
syncObj.wait();
}
}
catch(InterruptedException ie) { }
System.out.println("Main thread exiting...");
}
public void startBackgroundThread()
{
(new Thread(new Runnable()
{
#Override
public void run()
{
//Do something big...
System.out.println("Background Thread doing something big...");
//THEN HOW TO NOTIFY MAIN THREAD?
synchronized(syncObj)
{
System.out.println("Background Thread notifing...");
syncObj.notify();
}
System.out.println("Background Thread exiting...");
}
})).start();
}
}
and see this output
PS C:\Users\java> javac test.java
PS C:\Users\java> java test
Main thread waiting...
Background Thread doing something big...
Background Thread notifing...
Background Thread exiting...
Main thread exiting...

Just call notify()
public void run() {
try {
while ( true ) {
putMessage();
sleep( 1000 );
}
}
catch( InterruptedException e ) { }
}
private synchronized void putMessage() throws InterruptedException {
while ( messages.size() == MAXQUEUE )
wait();
messages.addElement( new java.util.Date().toString() );
notify();
}

You can't "notify the main thread".
The best approach is to use an ExecutorService, like this for example:
import java.util.concurrent.*;
// in main thread
ExecutorService executorService = Executors.newSingleThreadExecutor();
Future<?> future = executorService.submit(new Runnable() {
#Override
public void run() {
//Do something big...
}
});
future.get(); // blocks until the Runnable finishes
The classes are written specially to deal with asynchronous operations, and all the code in there is already written for you and bullet-proof.
Edit
If you don't want to block the main thread while waiting, wait within another thread:
final Future<?> future = executorService.submit(new Runnable() {
#Override
public void run() {
//Do something big...
}
});
new Thread(new Runnable() {
#Override
public void run() {
future.get(); // blocks until the other Runnable finishes
// Do something after the other runnable completes
}
}).start();

One thread notifying another thread is not a good way to do it. Its better to have 1 master thread that gives the slave thread work. The slave thread is always running and waits until it receives work. I recommend that you draw two columns and determine exactly where each thread needs to wait.
public void run()
{
//Do something big...
synchronized(this)
{
done = true;
}
}
Java includes libraries that make this really easy see ExecutorService and the following post
Producer/Consumer threads using a Queue

What is the simplest way in java to run many methods in separate threads and wait until all will be finished?

I want a method that runs 2 or more methods in separate threads. I want be sure that method won't finish before all threads are done.

The best approach is to utilize the Executor Service API to manage a thread pool instead of starting an open-ended number of threads on your own.
ExecutorService pool = Executors.newCachedThreadPool();
for (Runnable r : new Runnable[] {
new R() { void r() { myMethod1(); }},
new R() { void r() { myMethod2(); }},
})
pool.execute(r);
pool.shutdown();
pool.awaitTermination(60, TimeUnit.SECONDS);
abstract class R implements Runnable
public final void run() { r(); }
abstract void r();
}
Note that it is not advisable to insist on every method running in its own, separate thread. Threads are quite heavyweight (each allocating a complete call stack) and performance actually decreases as the thread count increases far beyond the number of available processor cores.

I prefer something like this:
public static void runParallel(Runnable... runnables) throws InterruptedException {
final CountDownLatch done = new CountDownLatch(runnables.length);
for (final Runnable r: runnables) {
new Thread(new Runnable() {
public void run() {
try {
r.run();
} finally {
done.countDown();
}
}
}).start();
}
done.await();
}
An advantage of this approach is that it also works with thread pool (i.e. you can replace new Thread(...).start() with executor.submit(...)).
Also it allows you to use pre-existing thread pool, unlike solutions based on awaitTermination() that force you to create new pools for each invocation.

My solution is
Function:
public void runParallel(Runnable... runnables) throws InterruptedException {
List<Thread> threads = new ArrayList<Thread>(runnables.length);
for (Runnable runnable :runnables) {
Thread th = new Thread(runnable);
threads.add(th);
th.start();
}
for (Thread th : threads) {
th.join();
}
Use:
runParallel(new Runnable() {
#Override
public void run() {
method1()
}
}, new Runnable() {
#Override
public void run() {
method2()
}
}
);
any better ideas? Maybe there is a shorter way that I'm not aware of ;)

Following the API given by damienix:
public void runParallel(Runnable... runnables) throws InterruptedException {
final ExecutorService pool = Executors.newFixedThreadPool(runnables.length);
for (Runnable runnable: runnables) {
pool.submit(runnable);
}
pool.shutdown();
pool.awaitTermination(1, TimeUnit.MINUTES);
}