Partitioned Spring Batch Step repeats the same successful slave StepExecutions

Question

Using Spring Batch 3.0.4.RELEASE.

I configure a job to use a partition step. The slave step uses chunk size 1. There are six threads in the task executor. I run this test with various grid sizes from six to hundreds. My grid size is the number of slave StepExecutions I expect == the number of ExecutionContexts created by my partitioner.

The result is always this:

The six threads pick up six different step executions and execute them successfully . Then the same six step executions run again and again in the same thread!

I notice that there is a loop in RepeatTemplate.executeInternal(...) that never ends. It keeps executing the same StepExecution just incrementing the version.

Here's the Java configuration code:

@Bean
@StepScope
public RapRequestItemReader rapReader(
        @Value("#{stepExecutionContext['" + RapJobConfig.LIST_OF_IDS_STEP_EXECUTION_CONTEXT_VAR + "']}") String listOfIds,
        final @Value("#{stepExecutionContext['" + RapJobConfig.TIME_STEP_EXECUTION_CONTEXT_VAR + "']}") String timeString) {
    final List<Asset> farms = Arrays.asList(listOfIds.split(",")).stream().map(intString -> assetDao.getById(Integer.valueOf(intString)))
            .collect(Collectors.toList());
    return new RapRequestItemReader(timeString, farms);
}

@Bean
public ItemProcessor<RapRequest, PullSuccess> rapProcessor() {
    return rapRequest -> {
        return rapPull.pull(rapRequest.timestamp, rapRequest.farms);
    };
}

@Bean
public TaskletStep rapStep1(StepBuilderFactory stepBuilderFactory, RapRequestItemReader rapReader) {
    return stepBuilderFactory.get(RAP_STEP_NAME)
            .<RapRequest, PullSuccess> chunk(RAP_STEP_CHUNK_SIZE)
            .reader(rapReader)
            .processor(rapProcessor())
            .writer(updateCoverageWriter)
            .build();
}

private RapFilePartitioner createRapFilePartitioner(RapParameter rapParameter) {
    RapFilePartitioner partitioner = new RapFilePartitioner(rapParameter, rapPull.getIncrementHours());
    return partitioner;
}

@Bean
public ThreadPoolTaskExecutor pullExecutor() {
    ThreadPoolTaskExecutor pullExecutor = new ThreadPoolTaskExecutor();
    pullExecutor.setCorePoolSize(weatherConfig.getNumberOfThreadsPerModelType());
    pullExecutor.setMaxPoolSize(weatherConfig.getNumberOfThreadsPerModelType());
    pullExecutor.setAllowCoreThreadTimeOut(true);
    return pullExecutor;
}

@Bean
@JobScope
public Step rapPartitionByTimestampStep(StepBuilderFactory stepBuilderFactory, @Value("#{jobParameters['config']}") String config,
        TaskletStep rapStep1) {
    RapParameter rapParameter = GsonHelper.fromJson(config, RapParameter.class);
    int gridSize = calculateGridSize(rapParameter);
    return stepBuilderFactory.get("rapPartitionByTimestampStep")
            .partitioner(rapStep1)
            .partitioner(RAP_STEP_NAME, createRapFilePartitioner(rapParameter))
            .taskExecutor(pullExecutor())
            .gridSize(gridSize)
            .build();
}

@Bean
public Job rapJob(JobBuilderFactory jobBuilderFactory, Step rapPartitionByTimestampStep) {
    return jobBuilderFactory.get(JOB_NAME)
            .start(rapPartitionByTimestampStep)
            .build();
}

Answer 1

Though it's hard to tell this from the question, the problem was in the reader. The ItemReader was never returning null.

In the design, a StepExecution was supposed to process only one item. However, after processing that item, the ItemReader was returning that same item again instead of returning null.

I fixed it by having the ItemReader return null the second time read is called.

A better design might be to use a TaskletStep instead of a ChunkStep.