Hadoop reducer not being called

Question

all

I have simple map/reduce implementation. Mapper is called and it does its job but reducer is never called.

Here is mapper:

static public class InteractionMap extends Mapper<LongWritable, Text, Text, InteractionWritable> {

    @Override
    protected void map(LongWritable offset, Text text, Context context) throws IOException, InterruptedException {
        System.out.println("mapper");
        String[] tokens = text.toString().split(",");
        for (int idx = 0; idx < tokens.length; idx++) {
            String sourceUser = tokens[1];
            String targetUser = tokens[2];
            int points = Integer.parseInt(tokens[4]);
            context.write(new Text(sourceUser), new InteractionWritable(targetUser, points));
            }
        }
    }
}

Here is my reducer:

static public class InteractionReduce extends Reducer<Text, InteractionWritable, Text, Text> {

    @Override
    protected void reduce(Text token, Iterable<InteractionWritable> counts, Context context) throws IOException, InterruptedException {
        System.out.println("REDUCER");
        Iterator<InteractionWritable> i = counts.iterator();
        while (i.hasNext()) {
            InteractionWritable interaction = i.next();
            context.write(token, new Text(token.toString() + " " + interaction.getTargetUser().toString() + " " + interaction.getPoints().get()));
        }
    }

}

And, here is configuration part:

@Override
public int run(String[] args) throws Exception {
    Configuration configuration = getConf();
    Job job = new Job(configuration, "Interaction Count");
    job.setJarByClass(InteractionMapReduce.class);
    job.setMapperClass(InteractionMap.class);
    job.setCombinerClass(InteractionReduce.class);
    job.setReducerClass(InteractionReduce.class);
    job.setInputFormatClass(TextInputFormat.class);
    job.setOutputFormatClass(TextOutputFormat.class);
    job.setOutputKeyClass(Text.class);
    job.setOutputValueClass(Text.class);
    FileInputFormat.addInputPath(job, new Path(args[0]));
    FileOutputFormat.setOutputPath(job, new Path(args[1]));
    return job.waitForCompletion(true) ? 0 : -1;
}

Does anyone have any idea why reducer is not being invoked?

Answer 1

Ok, it was my fault, as expected. Job configuration wasn't good. This is how it should look like:

Configuration configuration = getConf();

Job job = new Job(configuration, "Interaction Count");
job.setJarByClass(InteractionMapReduce.class);
job.setMapperClass(InteractionMap.class);
job.setReducerClass(InteractionReduce.class);
job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(InteractionWritable.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);

FileInputFormat.addInputPath(job, new Path(args[0]));
FileOutputFormat.setOutputPath(job, new Path(args[1]));

return job.waitForCompletion(true) ? 0 : -1;

The problem occurred because map and reduce phases have different output types. Job failed silently after invoking context.write method. So, what I had to add are these lines:

job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(InteractionWritable.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);

Answer 2

1. I hope the text in your Mapper method has some data.
2. Do you really need the Reducer to be the Combiner as well as the Reducer?

I always have one main class InteractionMapReduce and inside it I have the InteractionMap and the InteractionReduce class.

So while setting the Mapper and the Reducer class in the job, I set them like InteractionMapReduce.InteractionMap.class and InteractionMapReduce.InteractionReduce.class.

I do not know whether this would help you but you could try it.