I'm trying to create a method that executes a given task in a maximum amount of time. If it fails to finish in that time, it should be retried a number of times before giving up. It should also wait a number of seconds between each try. Here's what I've come up with and I'd like some critiques on my approach. Is their a simpler way to do this using the ScheduledExecutorService or is my way of doing this suffice?
public static <T> T execute(Callable<T> task, int tries, int waitTimeSeconds, int timeout)
throws InterruptedException, TimeoutException, Exception {
Exception lastThrown = null;
for (int i = 0; i < tries; i++) {
try {
final Future<T> future = new FutureTask<T>(task);
return future.get(timeout, TimeUnit.SECONDS);
} catch (TimeoutException ex) {
lastThrown = ex;
} catch (ExecutionException ex) {
lastThrown = (Exception) ex.getCause();
}
Thread.sleep(TimeUnit.SECONDS.toMillis(waitTimeSeconds));
}
if (lastThrown == null) {
lastThrown = new TimeoutException("Reached max tries without being caused by some exception. " + task.getClass());
}
throw lastThrown;
}
I think, but it's my opinion, that if you are scheduling network related tasks, you should not retry but eventually run them in parallel. I describe this other approach later.
Regarding your code, you should pass the task to an executor, or the FutureTask to a thread. It will not spawn a thread or execute by itself. If you have an executor (see ExecutorService), you don't even need a FutureTask, you can simply schedule it and obtain a callable.
So, given that you have an ExecutorService, you can call :
Future<T> future = yourExecutor.submit(task);
Future.get(timeout) will wait for that timeout and eventually return with TimeoutException even if the task has never started at all, for example if the Executor is already busy doing other work and cannot find a free thread. So, you could end up trying 5 times and waiting for seconds without ever giving the task a chance to run. This may or may not be what you expect, but usually it is not. Maybe you should wait for it to start before giving it a timeout.
Also, you should explicitly cancel the Future even if it throws TimeoutException, otherwise it may keep running, since nor documentation nor code says it will stop when a get with timeout fails.
Even if you cancel it, unless the Callable has been "properly written", it could keep running for some time. Nothing you can do it about it in this part of code, just keep in mind that no thread can "really stop" what another thread is doing in Java, and for good reasons.
However I suppose your tasks will mostly be network related, so it should react correctly to a thread interruption.
I usually use a different strategy is situations like this:
I would write public static T execute(Callable task, int maxTries, int timeout), so the task, max number of tries (potentially 1), max total timeout ("I want an answer in max 10 seconds, no matter how many times you try, 10 seconds or nothing")
I start spawning the task, giving it to an executor, and then call future.get(timeout/tries)
If I receive a result, return it. If I receive an exception, will try again (see later)
If however i get a timeout, I DON'T cancel the future, instead I save it in a list.
I check if too much time has passed, or too many retries. In that case I cancel all the futures in the list and throw exception, return null, whatever
Otherwise, I cycle, schedule the task again (in parallel with the first one).
See point 2
If I have not received a result, I check the future(s) in the list, maybe one of the previous spawned task managed to do it.
Assuming your tasks can be executed more than once (as I suppose they are, otherwise no way to retry), for network stuff I found this solution to work better.
Suppose your network is actually very busy, you ask for a network connection, giving 20 retries 2 seconds each. Since your network is busy, none of the 20 retries manages to get the connection in 2 seconds. However, a single execution lasting 40 seconds may manage to connect and receive data. It's like a person pressing f5 compulsively on a page when the net is slow, it will not do any good, since every time the browser has to start from the beginning.
Instead, I keep the various futures running, the first one that manages to get the data will return a result and the others will be stopped. If the first one hangs, the second one will work, or the third one maybe.
Comparing with a browser, is like opening another tab and retrying to load the page there without closing the first one. If the net is slow, the second one will take some time, but not stopping the first one, which will eventually load properly. If instead the first tab was hung, the second one will load rapidly. Whichever loads first, we can close the other tab.
The thread on which your execute is called will block for so much time. Not sure if this is correct for you. Basically , for these types of tasks , ScheduledExecutorService is best.You can schedule a task and specify the timings. Take a look at ScheduledThreadPoolExecutor
Related
I want to understand about the ideal value of timeout in awaitTermination() method for executorservice. How should we decide the ideal timeout? Does it harm if we use one day as the timeout?
boolean awaitTermination(long timeout, TimeUnit unit) throws InterruptedException;
suppose I am loading data from one database to another database. Sometimes it may slow down because of network issues and even the number of documents to upload may also increase. So it becomes a little difficult to judge the max time. So in that scenario, if I use one day as a timeout (Actually, I want to use awaitTermination() without a timeout from executorService side). What exactly are the disadvantages of this?
What is the downside of awaitTermination(Long.MAX_VALUE, TimeUnit.DAYS)?
because anyways, executorservice will terminate once the tasks are done
This depends on how long a task can consume.
How long in the worst case you longest task will take?
Lets say it will take in worst case 1 min. Then take 2 min.
You simply calculate worst case time plus a buffer.
You simply don't wont that the started job is canceled without reason because you can get a data loss.
If you are working e.g. on a queue with confirmation and you confirm only if your job is done, you can terminate immediately because on next start and after timeout your worker gets the task again. (But be carefully. If you servere restarts you can end up in a loop)
I have this simple code:
List<ActorRef> actors = new ArrayList<>();
for (int i = 0; i < ACTOR_COUNT; i++) {
actors.add(system.actorOf(...));
}
for (ActorRef actor : actors) {
system.scheduler().schedule(FiniteDuration.create(0, TimeUnit.MILLISECONDS),
FiniteDuration.create(1000, TimeUnit.MILLISECONDS), actor, "Run", system.dispatcher(), null);
}
It creates a number of actors and then creates a scheduler for each of them. Actors itself are responsible for querying MQ and then process a message.
When ACTOR_COUNT > 30, everything is good. But otherwise, we have a memory leak (instances of akka.dispatch.Envelopes with message "Run" are filling up and can't be garbage collected)
It's pretty weird, because when we have more actors, then we have more messages (1 per second for each of them) - but unexpectedly it STOPS filling up when there are more actors/messages.
Time interval (1000 ms) doesn't really affect the situation, it just make it slower or faster.
Could you please explain this behavior for me?
Thank you.
UPDATE
Here is a dummy actor, which can help to isolate a problem.
public class MessageQueueTestActor extends UntypedActor {
private static final Logger log = LoggerFactory.getLogger(MessageQueueTestActor.class);
#Override
public void onReceive(Object message) throws Exception {
Thread.sleep(3000);
}
}
The problem is reproduced with ACTOR_COUNT = 5. Now it's obvious that when actor sleep time > scheduler interval, envelopes are filling up. If I reduce sleep time from 3000ms to 500ms, the problem is gone.
But messages also become available for garbage collector if I increase the number of actors up to 30 (with the same sleep time = 3000ms). Why? Looks like something in Akka starts working differently after that threashold.
This is a «debug my code» question, not sure whether it should be here, but I'll answer in any case.
The scheduler does not enqueue the message into the actor’s mailbox itself, it uses the given dispatcher to do that. Since you block the threads in the default dispatcher and also use that to do the enqueueing, there is a point at which messages from the scheduler do not reach the mailboxes anymore (I assume that your default dispatcher has 30 threads). More correctly: they reach it one by one while the actors process up to five messages during each turn they get.
So, nothing is GC-ed, you just enqueue a different thing (Runnable) at a different place (default dispatcher). Your program will never work sustainably if the processing time is greater than the tick period.
Right now I have two threads running in my program. One constantly tries to read input from the user, and the other watches for a timeout. A timeout occurs if the user does not send any input in a given amount of time. The two threads look like this:
User input thread
while(true){
if(in.hasNextLine()){
processLine(in.nextLine());
timeLastRecieved = System.currentTimeMillis();
}
}
Timeout thread
while(true){
//Check for a timout
if(timeLastRecieved+timeoutDuration <= System.currentTimeMillis())
timeUserOut();
else{
//Sleep until it is possible for a timeout to occur
Thread.sleep((timeLastSent+timeoutDuration) - System.currentTimeMillis());
}
}
As of now I have these thread separated, but I could combine them like this...
while(true){
if(in.hasNextLine()){
processLine(in.nextLine());
timeLastRecieved = System.currentTimeMillis();
}
//Check for a timout
if(timeLastRecieved+timeoutDuration <= System.currentTimeMillis())
timeUserOut();
}
But I really don't need to check for a timeout that frequently. So should I combine the threads and check for a timeout too often, or should I have two threads. I am not as worried about performance as I am proper coding etiquette. If it means anything the timeout duration in something like 15 minutes long.
EDIT: Just want to point out that in the version with two thread I am sleeping, but in the combined version I never sleep the thread. This obviously causes the if statement that checks for a timeout to run more then necessary.
To summarize my comments: I don't think a separate thread to check for timeouts is necessary.
Reasons:
You'd need to share information like timeLastRecieved between them, which could be more complex than wanted (e.g. AFAIK in some cases access to long values is not atomic).
From your description it seems that polling for user input and timeout (no input provided in time) are closely related, thus the polling thread could check for the timeout as well. That doesn't mean it has to handle the timeout too, just reporting it somewhere or calling some timeout handler might be better design.
It is easier to read and understand since updating timeLastRecieved and checking for a timeout is handled in the same place.
Since there is no inter-thread communication nor coordination needed (there are no threads that need to communicate) it probably is more robust as well.
A few hints on checking for the timeout:
You should calculate the timeout threshold when you update timeLastReceived and then only check agains the current time instead of calculating it in every iteration.
You might want to calculate the timeout threshold before processing the input in order not to have it depend on the processing time as well.
Finally, there are alternative approaches like using java.util.Timer. Here you could simply schedule a timeout task which is executed when the timeout should occur. That task then would check if the timeout really happened and if not it just returns.
To handle new input before the timeout occured you could use at least two approaches:
Cancel the current timeout task, remove it from the timer and schedule a new one.
If there is already a scheduled timeout task then don't schedule a new one but wait for the current one to run. The current one then checks for the timeout and if none happened it schedules a new task (or itself) for the current anticipated timeout (note that this would require some inter-thread communcation so be careful here).
You need to have two threads - one waiting for data coming in through the InputStream / Reader, and one that's watching the time to see if the time elapsed as taken too long. The only way to do it with 1 thread would be to sleep for a segment of the timeout period and then poll for data periodically. But that's less efficient than having a separate thread dedicated to reading from your InputStream/Reader.
You may want to check out Timeout as a generic option for implementing a timeout
i just started to learn programming (2 weeks ago), and i am trying to make a bot for a game. In the main class of the bot, there are 3 methods that needs to be returned within 2second, or it will return null. I want to avoid returning null and return what it has calculate during 2sec instead.
public ArrayList<PlaceArmiesMove> getPlaceArmiesMoves(BotState state, Long timeOut){
ArrayList<PlaceArmiesMove> placeArmiesMoves = new ArrayList<PlaceArmiesMove>();
// caculations filling the ArrayList
return placeArmiesMoves;
}
what i want to do is after 2 second, returning placeArmiesMoves, wether the method finished running or not. I have read about guava SimpleTimeLimiter and callWithTimeout() but i am totally lost about how to use it (i read something about multithreading but i just don't understand what this is)
i would be incredibly grateful if someone could help me! thanks
Given a function like getPlaceArmiesMove, there are several techniques you might use to bound its execution time.
Trust the function to keep track of time itself
If the function runs a loop, it can check on every iteration whether the time has expired.
long startTime = System.currentTimeMillis()
for (;;) {
// do some work
long elapsed = System.currentTimeMillis() - startTime;
if (elapsed >= timeOut) {
break;
}
}
This technique is simple, but there is no guarantee it will complete before the timeout; it depends on the function and how granular you can make the work (of course, if it's too granular, you'll be spending more time testing if the timeout has expired than actually doing work).
Run the function in a thread, and ask it to stop
I'm not familiar with Guava, but this seems to be what SimpleTimeLimiter is doing. In Java, it isn't generally possible to forcibly stop a thread, though it is possible to ignore the thread after a timeout (the function will run to completion, but you've already used its partial result, and ignore the complete result that comes in too late). Guava says that it interrupts the thread if it has not returned before the timeout. This works only if your function is testing to see if it has been interrupted, much like the "trust your function" technique.
See this answer for an example on how to test if your thread has been interrupted. Note that some Java methods (like Thread.sleep) may throw InterruptedException if the thread is interrupted.
In the end, sprinkling checks for isInterrupted() all over your function won't be much different than sprinkling manual checks for the timeout. So running in a thread, you still must trust your function, but there may be nicer helpers available for that sort of thing (e.g. Guava).
Run the function in a separate process, and kill it
An example of how to do this is left as an exercise, but if you run your function in a separate process (or a thread in languages that support forcibly stopping threads, e.g. Erlang, Ruby, others), then you can use the operating system facilities to kill the process if it does not complete after a timeout.
Having that process return a partial result will be challenging. It could periodically send "work-in-progress" to the calling process over a pipe, or periodically save work to a file.
Use Java's Timer package , however this will require you to understand concepts such as threads and method overriding. Nevertheless, if this is what you require, the answer is quite similar to this question How to set a timer in java
This question already has answers here:
How can I abort a running JDBC transaction?
(4 answers)
Closed 5 years ago.
I have a program that continually polls the database for change in value of some field. It runs in the background and currently uses a while(true) and a sleep() method to set the interval. I am wondering if this is a good practice? And, what could be a more efficient way to implement this? The program is meant to run at all times.
Consequently, the only way to stop the program is by issuing a kill on the process ID. The program could be in the middle of a JDBC call. How could I go about terminating it more gracefully? I understand that the best option would be to devise some kind of exit strategy by using a flag that will be periodically checked by the thread. But, I am unable to think of a way/condition of changing the value of this flag. Any ideas?
I am wondering if this is a good practice?
No. It's not good. Sometimes, it's all you've got, but it's not good.
And, what could be a more efficient way to implement this?
How do things get into the database in the first place?
The best change is to fix programs that insert/update the database to make requests which go to the database and to your program. A JMS topic is good for this kind of thing.
The next best change is to add a trigger to the database to enqueue each insert/update event into a queue. The queue could feed a JMS topic (or queue) for processing by your program.
The fall-back plan is your polling loop.
Your polling loop, however, should not trivially do work. It should drop a message into a queue for some other JDBC process to work on. A termination request is another message that can be dropped into the JMS queue. When your program gets the termination message, it absolutely must be finished with the prior JDBC request and can stop gracefully.
Before doing any of this, look at ESB solutions. Sun's JCAPS or TIBCO already have this. An open source ESB like Mulesource or Jitterbit may already have this functionality already built and tested.
This is really too big an issue to answer completely in this format. Do yourself a favour and go buy Java Concurrency in Practice. There is no better resource for concurrency on the Java 5+ platform out there. There are whole chapters devoted to this subject.
On the subject of killing your process during a JDBC call, that should be fine. I believe there are issues with interrupting a JDBC call (in that you can't?) but that's a different issue.
As others have said, the fact that you have to poll is probably indicative of a deeper problem with the design of your system... but sometimes that's the way it goes, so...
If you'd like to handle "killing" the process a little more gracefully, you could install a shutdown hook which is called when you hit Ctrl+C:
volatile boolean stop = false;
Runtime.getRuntime().addShutdownHook(new Thread("shutdown thread") {
public void run() {
stop = true;
}
});
then periodically check the stop variable.
A more elegant solution is to wait on an event:
boolean stop = false;
final Object event = new Object();
Runtime.getRuntime().addShutdownHook(new Thread("shutdown thread") {
public void run() {
synchronized(event) {
stop = true;
event.notifyAll();
}
}
});
// ... and in your polling loop ...
synchronized(event) {
while(!stop) {
// ... do JDBC access ...
try {
// Wait 30 seconds, but break out as soon as the event is fired.
event.wait(30000);
}
catch(InterruptedException e) {
// Log a message and exit. Never ignore interrupted exception.
break;
}
}
}
Or something like that.
Note that a Timer (or similar) would be better in that you could at least reuse it and let it do with all of the details of sleeping, scheduling, exception handling, etc...
There are many reasons your app could die. Don't focus on just the one.
If it's even theoretically possible for your JDBC work to leave things in a half-correct state, then you have a bug you should fix. All of your DB work should be in a transaction. It should go or not go.
This is Java. Move your processing to a second thread. Now you can
Read from stdin in a loop. If someone types "QUIT", set the while flag to false and exit.
Create a AWT or Swing frame with a STOP button.
Pretend you are a Unix daemon and create a server socket. Wait for someone to open the socket and send "QUIT". (This has the added bonus that you can change the sleep to a select with timeout.)
There must be hundreds of variants on this.
Set up a signal handler for SIGTERM that sets a flag telling your loop to exit its next time through.
Regarding the question "The program could be in the middle of a JDBC call. How could I go about terminating it more gracefully?" - see How can I abort a running jdbc transaction?
Note that using a poll with sleep() is rarely the correct solution - implemented improperly, it can end up hogging CPU resources (the JVM thread-scheduler ends up spending inordinate amount of time sleeping and waking up the thread).
I‘ve created a Service class in my current company’s utility library for these kinds of problems:
public class Service implements Runnable {
private boolean shouldStop = false;
public synchronized stop() {
shouldStop = true;
notify();
}
private synchronized shouldStop() {
return shouldStop;
}
public void run() {
setUp();
while (!shouldStop()) {
doStuff();
sleep(60 * 1000);
}
}
private synchronized sleep(long delay) {
try {
wait(delay);
} catch (InterruptedException ie1) {
/* ignore. */
}
}
}
Of course this is far from complete but you should get the gist. This will enable you to simply call the stop() method when you want the program to stop and it will exit cleanly.
If that's your application and you can modify it, you can:
Make it read a file
Read for the value of a flag.
When you want to kill it, you just modify the file and the application will exit gracefully.
Not need to work it that harder that that.
You could make the field a compound value that includes (conceptually) a process-ID and a timestamp. [Better yet, use two or more fields.] Start a thread in the process that owns access to the field, and have it loop, sleeping and updating the timestamp. Then a polling process that is waiting to own access to the field can observe that the timestamp has not updated in some time T (which is much greater than the time of the updating loop's sleep interval) and assume that the previously-owning process has died.
But this is still prone to failure.
In other languages, I always try to use flock() calls to synchronize on a file. Not sure what the Java equivalent is. Get real concurrency if you at all possibly can.
I'm surprised nobody mentioned the interrupt mechanism implemented in Java. It's supposed to be a solution to the problem of stopping a thread. All other solutions have at least one flaw, that's why this mechanism is needed to be implemented in the Java concurrency library.
You can stop a thread by sending it an interrupt() message, but there are others ways that threads get interrupted. When this happens an InterruptedException is thrown. That's why you have to handle it when calling sleep() for example. That's where you can do cleanup and end gracefully, like closing the database connection.
Java9 has another "potential" answer to this: Thread.onSpinWait():
Indicates that the caller is momentarily unable to progress, until the occurrence of one or more actions on the part of other activities. By invoking this method within each iteration of a spin-wait loop construct, the calling thread indicates to the runtime that it is busy-waiting. The runtime may take action to improve the performance of invoking spin-wait loop constructions.
See JEP 285 for more details.
I think you should poll it with timertask instead.
My computer is running a while loop 1075566 times in 10 seconds.
Thats 107557 times in one second.
How often is it truly needed to poll it? A TimerTask runs at its fastest 1000 times in 1 second. You give it a parameter in int (miliseconds) as parameters. If you are content with that - that means you strain your cpu 108 times less with that task.
If you would be happy with polling once each second that is (108 * 1000). 108 000 times less straining. That also mean that you could check 108 000 values with the same cpu strain that you had with your one while loop - beause the you dont assign your cpu to check as often. Remember the cpu has a clock cycle. Mine is 3 600 000 000 hertz (cycles per second).
If your goal is to have it updated for a user - you can run a check each time the user logs in (or manually let him ask for an update) - that would practically not strain the cpu whatsoever.
You can also use thread.sleep(miliseconds); to lower the strain of your polling thread (as it wont be polling as often) you where doing.