Multiple Partition Consumers and Other Customizability

public interface PartitionConsumerConfiguration {

  // defines how/where the state is persisted and read from
  public byte[] readState();
  public void persistState(byte[] state);

  // defines a PartitionFilter, defining which partitions to entirely omit from the querying/processing
  public PartitionFilter getPartitionFilter();

  // defines an upper bound on how many partitions can be in the working set at any given time
  public int getMaxWorkingSetSize();

  // defines an upper bound on how many partitions can be processed by a run of a workflow instance
  public int getProcessingLimit();

  // called when processing of the retrieved partitions is successful
  public void onSuccess();
 
  // called when there is an error during processing of partitions
  public void onFailure();
}

public class PartitionConsumer {

  public Iterator<Partition> getPartitions(PartitionedFileSet pfs, PartitionConsumerConfiguration, partitionConsumerConfiguration) {
    PartitionConsumerState state = partitionConsumerConfiguration.readState();
    // get new partitions in the dataset that match a partition PartitionFilter
    Iterator<Partition> partitions =
      pfsDataset.consumePartitions(state.getPointer(), limit = getMaxWorkingSetSize() - state.getSize(), 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(getProcessingLimit());
    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() {
    PartitionConsumer partitionConsumer = new PartitionConsumer();
    PartitionConsumerConfiguration pcc = ... // Design of user-defined configuration - TBD
    PartitionConsumerResult partitionsToProcess;

    getContext().execute(new TxRunnable() {
      @Override
      public void run(DatasetContext context) throws Exception {
        PartitionedFileSet pfs = context.getDataset("customPFS"); 
        partitionsToProcess = partitionConsumer.getPartitions(pfs, pcc);
      }
    });

    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 {
        if (success) {
          // this will, by default, remove them from the working set because they have successfully been processed
          pcc.onSuccess(partitionsToProcess);
        } else {
          // this will, by default, leave the partitions in the working set, and simply mark them as NEW
          pcc.onFailure(partitionsToProcess); 
        }
      }
    });
  }
}