j-easy / easy-batch

The simple, stupid batch framework for Java
https://github.com/j-easy/easy-batch/wiki
MIT License
611 stars 199 forks source link

Writer will not handle records, if batchsize is not reached #413

Open thom2batki opened 1 year ago

thom2batki commented 1 year ago

When using BlockingQueues for processing incoming records, the writer-job will not execute, if the batchsize is not reached.

I have following Code:

public class Launcher {

    public static void main(String[] args) {

        BlockingQueue<Record<String>> processingQueue = new LinkedBlockingQueue<>();
        BlockingQueue<Record<String>> resultQueue = new LinkedBlockingQueue<>();

        Job readerJob = new JobBuilder<String, String>()
                .named("reader-job")
                .reader(new IterableRecordReader<>(Arrays.asList("one", "two", "three", "four")))
                .writer(new BlockingQueueRecordWriter<>(processingQueue))
                .batchSize(5)
                .build();

        Job processorJob = new JobBuilder<String, String>()
                .named("processor-job")
                .reader(new BlockingQueueRecordReader<>(processingQueue))
                .processor(new MyProcessor("processor-job"))
                .writer(new BlockingQueueRecordWriter<>(resultQueue))
                .batchSize(5)
                .build();

        Job writerJob = new JobBuilder<String, String>()
                .named("writer-job")
                .reader(new BlockingQueueRecordReader<>(resultQueue))
                .writer(new StandardOutputRecordWriter<>())
                .batchSize(5)
                .build();

        JobExecutor jobExecutor = new JobExecutor();
        jobExecutor.submitAll(readerJob, processorJob, writerJob);
        jobExecutor.shutdown();
    }

    static class MyProcessor implements RecordProcessor<String, String> {

        private String workerName;

        public MyProcessor(String workerName) {
            this.workerName = workerName;
        }

        @Override
        public Record processRecord(Record record) throws Exception {
            System.out.println(workerName + ": processing record " + record.getPayload());
            return record;
        }
    }
}

As you can see, batchsize is defined as 5 for all jobs and 4 records need to be processed. In this case, all records will be read and processed, but never be written.

There are basically three different behaviours, depending on batchsize and number of records:

  1. batchsize never reached -> records never written
  2. batchsize exacly reached -> all records written
  3. batchsize exceeded -> only common multiple of batchsize written, rest will never be written

It seems that the writer-job is closing the queue to early and not waiting for incoming records anymore. In earlier versions of easy-batch PoisonRecords were used to sync between jobs. Is there anything comparable in actual versions? I don't want to just define a _QUEUETIMEOUT , which does not feel like a clever solution for this problem.

Because of this fact, the provided easy-batch-tutorial is not working properly:

https://github.com/j-easy/easy-batch/blob/master/easy-batch-tutorials/src/main/java/org/jeasy/batch/tutorials/advanced/parallel/ForkJoin.java

Here, all records are getting processed by the workingJobs, but they will never be written to Std.out by the defined StandardOutputRecordWriter<>().

Can someone help me out with this issue or someone out there dealing with the same problem?

joytools commented 1 year ago

I am using a LOT parallel jobs and I found the same problem when migrating to the latest EasyBatch release. I prefer using PoisonRecords to achieve a finer control over the entire processing flow.

Here is how I solved the problem

  1. Downloaded previous EasyBatch version source code

  2. Rebuilt the following 5 classes from the original source code to a custom package

    org.mypackage.PoisonRecordFilter
    org.mypackage.PoisonrecordBroadcaster
    org.mypackage.PoisonRecord
    org.mypackage.PoisonBlockingQueueRecordReader
    org.mypackage.PoisonBlockingQueueRecordWriter
  3. Reintroduced the previous PoisonRecord based functionalities using the above classes in my package

Hope this may help.