Create the executor first.
You have several possibilites.
If I suppose that your tasks implement a simple interface to query their status (something like an enum with 'NeedReschedule' or 'Completed'), then implement a wrapper (implementing Runnable
) for your tasks which will take the task and the executor as instanciation parameters. This wrapper will run the task it is bound to, check its status afterwards, and if necessary reschedule a copy of itself in the executor before terminating.
Alternatively, you could use an execption mechanism to signal the wrapper that the task must be rescheduled.
This solution is simpler, in the sense that it doesn't require a particular interface for you task, so that simple Runnable
could be thrown in the system without trouble. However, exceptions incur more computation time (object construction, stack trace etc.).
Here's a possible implementation of the wrapper using the exception signaling mechanism.
You need to implement the RescheduleException
class extending Throwable
, which may be fired by the wrapped runnable (no need for a more specific interface for the task in this setup). You could also use a simple RuntimeException
as proposed in another answer, but you will have to test the message string to know if this is the exception you are waiting for.
public class TaskWrapper implements Runnable {
private final ExecutorService executor;
private final Runnable task;
public TaskWrapper(ExecutorService e, Runnable t){
executor = e;
task = t;
}
@Override
public void run() {
try {
task.run();
}
catch (RescheduleException e) {
executor.execute(this);
}
}
Here's a very simple application firing up 200 wrapped tasks randomly asking a reschedule.
class Task implements Runnable {
@Override
public void run(){
if (Maths.random() > 0.5)
throw new RescheduleException();
}
}
public class Main {
public static void main(String[] args){
ExecutorService executor = Executors.newFixedThreadPool(10);
int i = 200;
while(i--)
executor.execute(new TaskWrapper(executor, new Task());
}
}
You could also have a dedicated thread to monitor the other threads results (using a message queue) and reschedule if necessary, but you lose one thread, compared to the other solution.