...

// specifies which Dataset to write to and handles the delegation to the appropriate OutputFormat:
mos.write(String datasetName, KEY key, VALUE value);

Example Usage:

public void beforeSubmit(MapReduceContext context) throws Exception {
  context.addOutput("cleanCounts");
  context.addOutput("invalidCounts");
  // ...
}

public static class Counter extends Reducer<Text, IntWritable, byte[], Long> {
  private MultipleOutputs<byte[], Long> mos;

  @Override
  protected void setup(Context context) {
    mos = new MultipleOutputs<>(context);
  }

  @Override
  public void reduce(Text key, Iterable<IntWritable> values, Context context) {
    // do computation and output to the desired dataset
    if ( ... ) {
      mos.write("cleanCounts", key.getBytes(), val);
    } else {
      mos.write("invalidCounts", key.getBytes(), val);
    }
  }

  @Override
  protected void cleanup(Context context) {
    mos.close();
  }
}

Approach:

Take an approach similar to org.apache.hadoop.mapreduce.lib.output.MultipleOutputs.
The Datasets to be written to must be defined in advance, in the beforeSubmit of the MapReduce job.
In the mapper/reducer, the user specifies the name of the output Dataset, and our helper class (MultipleOutputs) determines the appropriate OutputFormat and configuration for writing.
The MapperWrapper and ReducerWrapper will be responsible for instantiating the MultipleOutputs class and setting it on the user's mapper/reducer in a similar fashion as Metrics are set. The MapperWrapper and ReducerWrapper will also be responsible for closing the MultipleOutputs object.

Questions:

Should we also allow specifying in the configure method? I suggest we deprecate the setting of output dataset from the configure method as it provides no utility over setting it in the beforeSubmit.
Naming of the MultipleOutputs class that we expose is up for change.
Does the MultipleOutputs class go into cdap-api? If so, how do we expose it to the user for instantiation (it has hadoop dependencies)? One possibility is to set the field in Mapper/Reducer-Wrapper.
Should we allow the user to write to non-Dataset files from our MultipleOutputs class? I suggest no for simplicity. What this will disallow is the ability to write to both a Dataset and non-Dataset files from the same MapReduce.

...