Implement a WordCount example by using a MapReduce pipeline - MaxCompute

This example shows how to count word frequencies using a MaxCompute MapReduce Pipeline job. The pipeline chains one mapper (TokenizerMapper) and two reducers (SumReducer and IdentityReducer), demonstrating how the Pipeline API connects multiple processing stages into a single job.

Prerequisites

Before you begin, ensure that you have:

Completed the environment setup described in Getting started

Set up test resources

1. Prepare the JAR package

Place mapreduce-examples.jar in the bin\data\resources directory of your MaxCompute client installation.

2. Create tables

Run the following SQL statements in the MaxCompute client:

CREATE TABLE wc_in (key STRING, value STRING);
CREATE TABLE wc_out(key STRING, cnt BIGINT);

3. Add the JAR package as a resource

add jar data\resources\mapreduce-examples.jar -f;

The -f flag overwrites an existing resource with the same name. Omit it the first time you add the JAR package.

4. Import test data

Use a Tunnel command to upload data.txt from the bin directory of the MaxCompute client into wc_in:

tunnel upload data.txt wc_in;

The data.txt file contains the following line:

hello,odps

Run the pipeline

Run the following command in the MaxCompute client:

jar -resources mapreduce-examples.jar -classpath data\resources\mapreduce-examples.jar
com.aliyun.odps.mapred.open.example.WordCountPipeline wc_in wc_out;

Verify the result

After the job completes, query wc_out to confirm the output:

SELECT * FROM wc_out;

The expected output is:

+------------+------------+
| key        | cnt        |
+------------+------------+
| hello      | 1          |
| odps       | 1          |
+------------+------------+

Sample code

For Project Object Model (POM) dependency configuration, see Precautions.

How the pipeline works

The pipeline chains three stages. Each stage transforms the data as follows:

wc_in (key STRING, value STRING) → TokenizerMapper → (word STRING, count BIGINT) → SumReducer → (word STRING, count BIGINT) → IdentityReducer → wc_out (key STRING, cnt BIGINT)

Class	Role
`TokenizerMapper`	Splits each record into tokens and emits `(word, 1)` for each token
`SumReducer`	Aggregates counts for each word key
`IdentityReducer`	Writes the final `(word, count)` pairs to the output table

The two-reducer design separates aggregation from output formatting. SumReducer accumulates totals across all mapper outputs. IdentityReducer then maps the intermediate schema to the wc_out table schema, keeping the two concerns independent and making each reducer easier to test and replace.

If you omit OutputKeySortColumns, PartitionColumns, and OutputGroupingColumns for a mapper during pipeline construction, the framework uses the mapper's OutputKey as the default value for all three parameters.

Java code

package com.aliyun.odps.mapred.open.example;
import java.io.IOException;
import java.util.Iterator;
import com.aliyun.odps.Column;
import com.aliyun.odps.OdpsException;
import com.aliyun.odps.OdpsType;
import com.aliyun.odps.data.Record;
import com.aliyun.odps.data.TableInfo;
import com.aliyun.odps.mapred.Job;
import com.aliyun.odps.mapred.MapperBase;
import com.aliyun.odps.mapred.ReducerBase;
import com.aliyun.odps.pipeline.Pipeline;
public class WordCountPipelineTest {
    public static class TokenizerMapper extends MapperBase {
        Record word;
        Record one;
        @Override
            public void setup(TaskContext context) throws IOException {
            word = context.createMapOutputKeyRecord();
            one = context.createMapOutputValueRecord();
            one.setBigint(0, 1L);
        }
        @Override
            public void map(long recordNum, Record record, TaskContext context)
            throws IOException {
            for (int i = 0; i < record.getColumnCount(); i++) {
                String[] words = record.get(i).toString().split("\\s+");
                for (String w : words) {
                    word.setString(0, w);
                    context.write(word, one);
                }
            }
        }
    }
    public static class SumReducer extends ReducerBase {
        private Record value;
        @Override
            public void setup(TaskContext context) throws IOException {
            value = context.createOutputValueRecord();
        }
        @Override
            public void reduce(Record key, Iterator<Record> values, TaskContext context)
            throws IOException {
            long count = 0;
            while (values.hasNext()) {
                Record val = values.next();
                count += (Long) val.get(0);
            }
            value.set(0, count);
            context.write(key, value);
        }
    }
    public static class IdentityReducer extends ReducerBase {
        private Record result;
        @Override
            public void setup(TaskContext context) throws IOException {
            result = context.createOutputRecord();
        }
        @Override
            public void reduce(Record key, Iterator<Record> values, TaskContext context)
            throws IOException {
            while (values.hasNext()) {
                result.set(0, key.get(0));
                result.set(1, values.next().get(0));
                context.write(result);
            }
        }
    }
    public static void main(String[] args) throws OdpsException {
        if (args.length != 2) {
            System.err.println("Usage: WordCountPipeline <in_table> <out_table>");
            System.exit(2);
        }
        Job job = new Job();
        /** During pipeline construction, if you do not specify OutputKeySortColumns, PartitionColumns, and OutputGroupingColumns for a mapper, the framework uses OutputKey of the mapper as the default values of these parameters.
         */
        Pipeline pipeline = Pipeline.builder()
            .addMapper(TokenizerMapper.class)
            .setOutputKeySchema(
            new Column[] { new Column("word", OdpsType.STRING) })
            .setOutputValueSchema(
            new Column[] { new Column("count", OdpsType.BIGINT) })
            .setOutputKeySortColumns(new String[] { "word" })
            .setPartitionColumns(new String[] { "word" })
            .setOutputGroupingColumns(new String[] { "word" })
            .addReducer(SumReducer.class)
            .setOutputKeySchema(
            new Column[] { new Column("word", OdpsType.STRING) })
            .setOutputValueSchema(
            new Column[] { new Column("count", OdpsType.BIGINT)})
            .addReducer(IdentityReducer.class).createPipeline();
        /** Add the pipeline to jobconf. If you want to configure a combiner, use jobconf. */
        job.setPipeline(pipeline);
        /** Configure the input and output tables. */
        job.addInput(TableInfo.builder().tableName(args[0]).build());
        job.addOutput(TableInfo.builder().tableName(args[1]).build());
        /** Submit the job and wait for it to complete. */
        job.submit();
        job.waitForCompletion();
        System.exit(job.isSuccessful() == true ? 0 : 1);
    }
}

Data flow walk-through

With input hello,odps, the Tunnel command loads the data as a single record into wc_in, storing hello in the key column and odps in the value column.

TokenizerMapper iterates over all columns of each record using record.getColumnCount(). For each column value, it calls split("\\s+") and emits one (word, 1) pair per token. Because hello and odps each contain no whitespace, each is treated as a single token. The mapper emits:

(hello, 1)
(odps, 1)

SumReducer receives all values for each word key and accumulates the count. Each word appears once, so it emits:

(hello, 1)
(odps, 1)

IdentityReducer reads the (word, count) pairs from SumReducer and writes them to the wc_out output table, mapping column positions to match the (key STRING, cnt BIGINT) schema.

Final output in wc_out:

| hello      | 1          |
| odps       | 1          |