Page Comparison

/**
 * Batch Sink that writes to a FileSet in text format.
 * Each record will be written as a single line, with record fields separated by a configurable separator.
 *
 * StructuredRecord is the first parameter because that is the input to the sink.
 * The second and third parameters are the key and value expected by Hadoop's {@link TextOutputFormat}.
 */
@Plugin(type = BatchSink.PLUGIN_TYPE)
@Name(TextFileSetSink.NAME)
@Description("Writes to a FileSet in text format.")
public class TextFileSetSink extends BatchSink<StructuredRecord, NullWritable, Text> {
  public static final String NAME = "TextFileSet";
  private final Conf config;

  /**
   * Config properties for the plugin.
   */
  public static class Conf extends PluginConfig {
    public static final String FILESET_NAME = "fileSetName";
    public static final String OUTPUT_DIR = "outputDir";
    public static final String FIELD_SEPARATOR = "fieldSeparator";

    // The name annotation tells CDAP what the property name is. It is optional, and defaults to the variable name.
    // Note:  only primitives (including boxed types) and string are the types that are supported
    @Name(FILESET_NAME)
    @Description("The name of the FileSet to write to.")
    private String fileSetName;

    // Macro enabled properties can be set to a placeholder value ${key} when the pipeline is deployed.
    // At runtime, the value for 'key' can be given and substituted in.
    @Macro
    @Name(OUTPUT_DIR)
    @Description("The FileSet directory to write to.")
    private String outputDir;

    @Nullable
    @Name(FIELD_SEPARATOR)
    @Description("The separator to use to join input record fields together. Defaults to ','.")
    private String fieldSeparator;

    // Use a no-args constructor to set field defaults.
    public Conf() {
      fileSetName = "";
      fieldSeparator = ",";
    }
  }

  // CDAP will pass in a config with its fields populated based on the configuration given when creating the pipeline.
  public TextFileSetSink(Conf config) {
    this.config = config;
  }

  // configurePipeline is called exactly once when the pipeline is being created.
  // Any static configuration should be performed here.
  @Override
  public void configurePipeline(PipelineConfigurer pipelineConfigurer) {
    // create the FileSet here.
    pipelineConfigurer.createDataset(config.fileSetName,
                                     FileSet.class,
                                     FileSetProperties.builder()
                                       .setInputFormat(TextInputFormat.class)
                                       .setOutputFormat(TextOutputFormat.class)
                                       .setEnableExploreOnCreate(true)
                                       .setExploreFormat("text")
                                       .setExploreSchema("text string")
                                       .build()
    );
  }

  // prepareRun is called before every pipeline run, and is used to configure what the input should be,
  // as well as any arguments the input should use. It is called by the client that is submitting the batch job.
  @Override
  public void prepareRun(BatchSinkContext context) throws Exception {
    Map<String, String> arguments = new HashMap<>();
    FileSetArguments.setOutputPath(arguments, config.outputDir);
    context.addOutput(Output.ofDataset(config.fileSetName, arguments));
  }

  // onRunFinish is called at the end of the pipeline run by the client that submitted the batch job.
  @Override
  public void onRunFinish(boolean succeeded, BatchSinkContext context) {
    // perform any actions that should happen at the end of the run.
  }

  // initialize is called by each job executor before any call to transform is made.
  // This occurs at the start of the batch job run, after the job has been successfully submitted.
  // For example, if mapreduce is the execution engine, each mapper will call initialize at the start of the program.
  @Override
  public void initialize(BatchRuntimeContext context) throws Exception {
    super.initialize(context);
    // create any resources required by transform()
  }

  // destroy is called by each job executor at the end of its life.
  // For example, if mapreduce is the execution engine, each mapper will call destroy at the end of the program.
  @Override
  public void destroy() {
    // clean up any resources created by initialize
  }

  @Override
  public void transform(StructuredRecord input, Emitter<KeyValue<NullWritable, Text>> emitter) throws Exception {
    StringBuilder joinedFields = new StringBuilder();
    Iterator<Schema.Field> fieldIter = input.getSchema().getFields().iterator();
    if (!fieldIter.hasNext()) {
      // shouldn't happen
      return;
    }

    Object val = input.get(fieldIter.next().getName());
    if (val != null) {
      joinedFields.append(val);
    }
    while (fieldIter.hasNext()) {
      String fieldName = fieldIter.next().getName();
      joinedFields.append(config.fieldSeparator);
      val = input.get(fieldName);
      if (val != null) {
        joinedFields.append(val);
      }
    }
    emitter.emit(new KeyValue<>(NullWritable.get(), new Text(joinedFields.toString())));
  }

}

/**
 * Aggregator that counts how many times each word appears in records input to the aggregator.
 */
@Plugin(type = BatchAggregator.PLUGIN_TYPE)
@Name(WordCountAggregator.NAME)
@Description("Counts how many times each word appears in all records input to the aggregator.")
public class WordCountAggregator extends BatchAggregator<String, StructuredRecord, StructuredRecord> {
  public static final String NAME = "WordCount";
  public static final Schema OUTPUT_SCHEMA = Schema.recordOf(
    "wordCount",
    Schema.Field.of("word", Schema.of(Schema.Type.STRING)),
    Schema.Field.of("count", Schema.of(Schema.Type.LONG))
  );
  private static final Pattern WHITESPACE = Pattern.compile("\\s");
  private final Conf config;

  /**
   * Config properties for the plugin.
   */
  public static class Conf extends PluginConfig {
    @Description("The field from the input records containing the words to count.")
    private String field;
  }

  public WordCountAggregator(Conf config) {
    this.config = config;
  }

  @Override
  public void configurePipeline(PipelineConfigurer pipelineConfigurer) {
    // any static configuration validation should happen here.
    // We will check that the field is in the input schema and is of type string.
    Schema inputSchema = pipelineConfigurer.getStageConfigurer().getInputSchema();
    // a null input schema means its unknown until runtime, or its not constant
    if (inputSchema != null) {
      // if the input schema is constant and known at configure time, check that the input field exists and is a string.
      Schema.Field inputField = inputSchema.getField(config.field);
      if (inputField == null) {
        throw new IllegalArgumentException(
          String.format("Field '%s' does not exist in input schema %s.", config.field, inputSchema));
      }
      Schema fieldSchema = inputField.getSchema();
      Schema.Type fieldType = fieldSchema.isNullable() ? fieldSchema.getNonNullable().getType() : fieldSchema.getType();
      if (fieldType != Schema.Type.STRING) {
        throw new IllegalArgumentException(
          String.format("Field '%s' is of illegal type %s. Must be of type %s.",
                        config.field, fieldType, Schema.Type.STRING));
      }
    }
    // set the output schema so downstream stages will know their input schema.
    pipelineConfigurer.getStageConfigurer().setOutputSchema(OUTPUT_SCHEMA);
  }

  @Override
  public void groupBy(StructuredRecord input, Emitter<String> groupKeyEmitter) throws Exception {
    String val = input.get(config.field);
    if (val == null) {
      return;
    }

    for (String word : WHITESPACE.split(val)) {
      groupKeyEmitter.emit(word);
    }
  }

  @Override
  public void aggregate(String groupKey, Iterator<StructuredRecord> groupValues,
                        Emitter<StructuredRecord> emitter) throws Exception {
    long count = 0;
    while (groupValues.hasNext()) {
      groupValues.next();
      count++;
    }
    emitter.emit(StructuredRecord.builder(OUTPUT_SCHEMA).set("word", groupKey).set("count", count).build());
  }
}

/**
 * SparkCompute plugin that counts how many times each word appears in records input to the compute stage.
 */
@Plugin(type = SparkCompute.PLUGIN_TYPE)
@Name(WordCountCompute.NAME)
@Description("Counts how many times each word appears in all records input to the aggregator.")
public class WordCountCompute extends SparkCompute<StructuredRecord, StructuredRecord> {
  public static final String NAME = "WordCount";
  public static final Schema OUTPUT_SCHEMA = Schema.recordOf(
    "wordCount",
    Schema.Field.of("word", Schema.of(Schema.Type.STRING)),
    Schema.Field.of("count", Schema.of(Schema.Type.LONG))
  );
  private final Conf config;

  /**
   * Config properties for the plugin.
   */
  public static class Conf extends PluginConfig {
    @Description("The field from the input records containing the words to count.")
    private String field;
  }

  public WordCountCompute(Conf config) {
    this.config = config;
  }

  @Override
  public void configurePipeline(PipelineConfigurer pipelineConfigurer) {
    // any static configuration validation should happen here.
    // We will check that the field is in the input schema and is of type string.
    Schema inputSchema = pipelineConfigurer.getStageConfigurer().getInputSchema();
    if (inputSchema != null) {
      WordCount wordCount = new WordCount(config.field);
      wordCount.validateSchema(inputSchema);
    }
    // set the output schema so downstream stages will know their input schema.
    pipelineConfigurer.getStageConfigurer().setOutputSchema(OUTPUT_SCHEMA);
  }

  @Override
  public JavaRDD<StructuredRecord> transform(SparkExecutionPluginContext sparkExecutionPluginContext,
                                             JavaRDD<StructuredRecord> javaRDD) throws Exception {
    WordCount wordCount = new WordCount(config.field);
    return wordCount.countWords(javaRDD)
      .flatMap(new FlatMapFunction<Tuple2<String, Long>, StructuredRecord>() {
        @Override
        public Iterable<StructuredRecord> call(Tuple2<String, Long> stringLongTuple2) throws Exception {
          List<StructuredRecord> output = new ArrayList<>();
          output.add(StructuredRecord.builder(OUTPUT_SCHEMA)
                       .set("word", stringLongTuple2._1())
                       .set("count", stringLongTuple2._2())
                       .build());
          return output;
        }
      });
  }
}

/**
 * SparkSink plugin that counts how many times each word appears in records input to it and stores the result in
 * a KeyValueTable.
 */
@Plugin(type = SparkSink.PLUGIN_TYPE)
@Name(WordCountSink.NAME)
@Description("Counts how many times each word appears in all records input to the aggregator.")
public class WordCountSink extends SparkSink<StructuredRecord> {
  public static final String NAME = "WordCount";
  private final Conf config;

  /**
   * Config properties for the plugin.
   */
  public static class Conf extends PluginConfig {
    @Description("The field from the input records containing the words to count.")
    private String field;

    @Description("The name of the KeyValueTable to write to.")
    private String tableName;
  }

  public WordCountSink(Conf config) {
    this.config = config;
  }

  @Override
  public void configurePipeline(PipelineConfigurer pipelineConfigurer) {
    // any static configuration validation should happen here.
    // We will check that the field is in the input schema and is of type string.
    Schema inputSchema = pipelineConfigurer.getStageConfigurer().getInputSchema();
    if (inputSchema != null) {
      WordCount wordCount = new WordCount(config.field);
      wordCount.validateSchema(inputSchema);
    }
    pipelineConfigurer.createDataset(config.tableName, KeyValueTable.class, DatasetProperties.EMPTY);
  }

  @Override
  public void prepareRun(SparkPluginContext sparkPluginContext) throws Exception {
    // no-op
  }

  @Override
  public void run(SparkExecutionPluginContext sparkExecutionPluginContext,
                  JavaRDD<StructuredRecord> javaRDD) throws Exception {
    WordCount wordCount = new WordCount(config.field);
    JavaPairRDD outputRDD = wordCount.countWords(javaRDD)
      .mapToPair(new PairFunction<Tuple2<String, Long>, byte[], byte[]>() {
        @Override
        public Tuple2<byte[], byte[]> call(Tuple2<String, Long> stringLongTuple2) throws Exception {
          return new Tuple2<>(Bytes.toBytes(stringLongTuple2._1()), Bytes.toBytes(stringLongTuple2._2()));
        }
      });
    sparkExecutionPluginContext.saveAsDataset(outputRDD, config.tableName);
  }
}