Multiple Partition Consumers and Other Customizability

/**
 * Configuration parameters to be used by a {@link PartitionConsumer}.
 */
public class ConsumerConfiguration {

  ...


  /**
   * @return A predicate to be applied on {@link PartitionDetail}s to determine which partitions to include in the
   *         partition consumption.
   */
  public Predicate<PartitionDetail> getPartitionPredicate() {
    return partitionPredicate;
  }

  /**
   * @return An upper bound on the size of the working set of partitions that get serialized as part of the consumer's
   *         state.
   */
  public int getMaxWorkingSetSize() {
    return maxWorkingSetSize;
  }

  /**
   * Defines an expiration timeout, in seconds, of IN_PROGRESS partitions
   * @return number of seconds that a partition can be in progress before it is deemed failed.
   */
  public long getTimeout() {
    return timeout;
  }

/**
 * Responsible for determining whether to accept more Partitions.
 * Generally used to dynamically determine the number of partitions to consume.
 */
public interface PartitionAccepter {
  public enum Response {
    CONTINUE, STOP;
  }

  public Response accept(PartitionDetail partitionDetail);
}

public class PartitionConsumer {

  // move params to constructor?
  public Iterator<Partition> getPartitions(PartitionedFileSet pfs, PartitionConsumerConfiguration, partitionConsumerConfiguration) {
    PartitionConsumerState state = partitionConsumerConfiguration.readState();
    // get new partitions in the dataset that match a partition PartitionFilter
    int limit = getMaxWorkingSetSize() - state.getSize();
    Iterator<Partition> partitions = pfsDataset.consumePartitions(state.getPointer(), limit, getPartitionFilter());
    // add these new partitions into the working set.
    state.addPartitions(partitions);

    // get a number of partitions from the working set, mark them IN_PROGRESS, and return them.
    Iterator<Partition> partitionsToProcess = state.getPartitions(getConsumingAccepter());
    partitionConsumerConfiguration.persistState(state);
    // need to commit this transaction now, so that other instances of the processing entity see these partitions as IN_PROGRESS.
    return partitionsToProcess;
  }
} 

/**
 * Worker which runs one iteration of processing new partitions.
 * Note: As we see frequent patterns, we can abstract the common parts of the example usage into CDAP code.
 */
public class CustomWorker extends AbstractWorker {

  @Override
  public void run() {
    TransactionalPartitionConsumer partitionConsumer =
      new TransactionalPartitionConsumer(getContext(), "lines",
                                         new KVTableStatePersistor("consumingState", "state.key"));
    final List<PartitionDetail> partitions = partitionConsumer.consumePartitions();
    
    boolean success = true;
    try {
      // process partitionsToProcess
    } catch (Throwable t) {
      success = false;
      // log the error
    }
 
    getContext().execute(new TxRunnable() {
      @Override
      public void run(DatasetContext context) throws Exception {
        partitionConsumer.onFinish(partitionsToProcess, success);
      }
    });
  }
}

/**
 * MapReduce job which incrementally processes new partitions of a {@link PartitionedFileSet}.
 */
public class DataCleansingMapReduce extends AbstractMapReduce {
 
  private PartitionBatchInput.BatchPartitionCommitter partitionCommitter;

  @Override
  public void beforeSubmit(MapReduceContext context) throws Exception {
    partitionCommitter =
      PartitionBatchInput.setInput(context, DataCleansing.RAW_RECORDS,
                                   new KVTableStatePersistor(DataCleansing.CONSUMING_STATE, "state.key"));
    // set the output dataset as well as mapper and reducer classes
    ...
  }

  @Override
  public void onFinish(boolean succeeded, MapReduceContext context) throws Exception {
    partitionCommitter.onFinish(succeeded);
  }
 
  // define the mapper and reducer classes
  ...
}