Tailer synchronization after rolling

[Mickael Marrache]

Hi,

I have multiple threads concurrently appending to the same queue. Every thread appends according to the following code snippet:

ExcerptAppender appender = queue.createAppender();
appender.writeMap(msg.toMap());
appender.close();

As you can see, every time a thread needs to write a message, it creates an appender (which is stored in thread local), writes the message an closes the appender (I'm not sure if it's needed to close the appender after each write?)

Also, I have one tailer created by a separate application as follows:

ExcerptTailer reader = queue.createTailer();

Map<String, Object> msg;
while(running) {
       while((msg = reader.readMap()) != null) {
            //Add the message to some list for later processing
       }

       //Process the bulk of messages that have just been read
}

reader.close();

Note that the tailer is kept open until the application is stopped.

Using the following code, I see that my tailer doesn't always synchronize with the cycles. I got multiple times in the situation where the tailer stays on the same cycle while the appenders continuously appends to new files (new cycles).

Any idea?

Thanks,

Mickael

[Mickael Marrache]

I forgot to mention that I use Chronicle Queue v4.3.2.

[Peter Lawrey]

Could you provide a unit test which demonstrates this. We have a test where this passes, but we are probably doing something different.

--
You received this message because you are subscribed to the Google Groups "Chronicle" group.
To unsubscribe from this group and stop receiving emails from it, send an email to java-chronicl...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Mickael Marrache]

I made an additional test where my consumer and producers applications runs concurrently without debug.

The producer app writes a map message every 5 seconds and the rolling is configured to be minutely.

I can see the files rolling. On the other hand, the consumer app is able to process messages correctly moving between one cycle to the next.

However, when I do this kind of execution in debug mode using my IDE, I get into the case where the tailer is stuck and the readMap() method continuously returns null.

I will try to write a unit test that reproduces the issue.

[Peter Lawrey]

We have had problems in the past where the debugger interferes with the running of the problem.  This is still a bug, though it is likely to have a different cause.

[Mickael Marrache]

Here is the test you can use:

public class ChronicleRollingIssueTest {

 @Test
 public void test() throws Exception {
     final ChronicleQueue writeQueue = ChronicleQueueBuilder
       .single("your-path")
       .rollCycle(RollCycles.MINUTELY).build();
 
     final ChronicleQueue readQueue = ChronicleQueueBuilder
       .single("your-path")
       .rollCycle(RollCycles.MINUTELY).build();
 
     Runnable appendRunnable = new Runnable() {
         @Override
         public void run() {
             ExcerptAppender appender = writeQueue.createAppender();
             Map<String, Object> map = new HashMap<String, Object>();
             map.put("k1", "v1");
             appender.writeMap(map);
             appender.close();
         }
     };
 
     new Thread(appendRunnable, "appender t1").start();
     Thread.sleep(3000);
     new Thread(appendRunnable, "appender t2").start();
     Thread.sleep(3000);
 
     ExcerptTailer tailer = readQueue.createTailer();
 
     Map<String, Object> map;
     while(!Thread.currentThread().isInterrupted()) {
        while((map = tailer.readMap()) != null) {
           //The test enters here for every written message until rolling is performed where readMap() continuously returns null
           System.out.println("We just read a map");
        }
       
        new Thread(appendRunnable, "appender ti").start();
        Thread.sleep(3000);
     }
 
     tailer.close();
  }
}

I've tried running the test multiple times and I'm able to reproduce the issue.

Le dimanche 8 mai 2016 16:25:09 UTC+3, Peter Lawrey a écrit :

[Mickael Marrache]

I also tried running the test from the IDE without debug mode, and I still get the same issue. I don't think it's related to the debugger.

[Peter Lawrey]

Ok, Thank you. I will look into it tomorrow.

[Mickael Marrache]

I can add that I deployed my app on a Linux server (CentOS 6.7) and I didn't encounter this issue until now.

It may be a Windows related issue... I will perform more testing and let you know.

[Peter Lawrey]

Even though I believe it's windows related, we should still fix it.

[Mickael Marrache]

I agree. I just intended to provide you more information hoping it will be easier for you to find the issue :)

[Peter Lawrey]

We really appreciate the feedback. :)

[Peter Lawrey]

Hello Mickael,

The API was broken as you shouldn't be able to close() on the Appender or Tailer (but you could) So I am restructuring the API so you can't call it this way.  I have rewritten the test for 64 threads and SECONDLY updates and I am testing it now.  I have a couple of meetings but I hope to fix it today.

I have added a test, which is showing a different problem occasionally on start up. 

https://github.com/OpenHFT/Chronicle-Queue/blob/master/src/test/java/net/openhft/chronicle/queue/ChronicleRollingIssueTest.java

Regards,

   Peter.