Write To Aerospike via mapPartitions

      Write to Aerospike from spark via MapPartitions

      Problem Statement :  Data from HDFS needs be read from spark and saved in Aerospike. One needs to use mapPartition transformation to achieve the same.





package com.big.data.spark;

import com.aerospike.client.AerospikeClient;
import com.aerospike.client.Bin;
import com.aerospike.client.Key;
import com.aerospike.client.policy.WritePolicy;
import com.big.data.avro.schema.Employee;
import com.databricks.spark.avro.SchemaConverters;
import org.apache.commons.io.IOUtils;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.conf.Configured;
import org.apache.hadoop.util.Tool;
import org.apache.hadoop.util.ToolRunner;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.FlatMapFunction;
import org.apache.spark.sql.DataFrame;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SQLContext;
import org.apache.spark.sql.types.StructType;

import java.io.Closeable;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Iterator;
import java.util.List;

public class ReadFromHdfsWriteToAerospikeSparkMapPartition extends Configured implements Tool, Closeable {

    public static final String INPUT_PATH = "spark.input.path";
    public static final String OUTPUT_PATH = "spark.output.path";
    public static final String IS_RUN_LOCALLY = "spark.is.run.local";
    public static final String DEFAULT_FS = "spark.default.fs";
    public static final String NUM_PARTITIONS = "spark.num.partitions";

    //Aerospike related properties
    public static final String AEROSPIKE_HOSTNAME = "aerospike.host.name";
    public static final String AEROSPIKE_PORT = "aerospike.port";
    public static final String AEROSPIKE_NAMESPACE = "aerospike.name.space";
    public static final String AEROSPIKE_SETNAME = "aerospike.set.name";

    // For Dem key is emp_id and value is emp_name
    public static final String KEY_NAME = "avro.key.name";
    public static final String VALUE_NAME = "avro.value.name";

    private SQLContext sqlContext;
    private JavaSparkContext javaSparkContext;

    protected <T> JavaSparkContext getJavaSparkContext(final boolean isRunLocal,
                                                       final String defaultFs,
                                                       final Class<T> tClass) {
        final SparkConf sparkConf = new SparkConf()
                //Set spark conf here ,
               // After one gets spark context you can set hadoop configuration for InputFormats
                .set("spark.serializer", "org.apache.spark.serializer.KryoSerializer");

        if (isRunLocal) {

        final JavaSparkContext sparkContext = new JavaSparkContext(sparkConf);

        if (defaultFs != null) {
            sparkContext.hadoopConfiguration().set("fs.defaultFS", defaultFs);

        return sparkContext;

    public int run(String[] args) throws Exception {

        //The arguments passed has been split into Key value by ToolRunner
        Configuration conf = getConf();
        String inputPath = conf.get(INPUT_PATH);
        String outputPath = conf.get(OUTPUT_PATH);
        String aerospikeHostname = conf.get(AEROSPIKE_HOSTNAME);
        int aerospikePort = conf.getInt(AEROSPIKE_PORT, 3000);
        String namespace = conf.get(AEROSPIKE_NAMESPACE);
        String setName = conf.get(AEROSPIKE_SETNAME);

        String keyName = conf.get(KEY_NAME);
        String valueName = conf.get(VALUE_NAME);

        //Get spark context, This is the central context , which can be wrapped in Any Other context
        javaSparkContext = getJavaSparkContext(conf.getBoolean(IS_RUN_LOCALLY, Boolean.FALSE),
                                                          conf.get(DEFAULT_FS), this.getClass());
        sqlContext = new SQLContext(javaSparkContext);

        // No input path has been read, no job has not been started yet .
        //To set any configuration use javaSparkContext.hadoopConfiguration().set(Key,value);
        // To set any custom inputformat use javaSparkContext.newAPIHadoopFile() and get a RDD

        // Avro schema to StructType conversion
        final StructType outPutSchemaStructType = (StructType) SchemaConverters
        final StructType inputSchema = (StructType) SchemaConverters

        // read data from parquetfile, the schema of the data is taken from the avro schema
        DataFrame inputDf = sqlContext.read().schema(inputSchema).parquet(inputPath);

        // convert DataFrame into JavaRDD
        // the rows read from the parquetfile is converted into a Row object .
        // Row has same schema as that of the parquet file roe
        JavaRDD<Row> rowJavaRDD = inputDf.javaRDD();

        // Data read from parquet has same schema as that of avro (Empoyee Avro). 
        // Key is employeeId and value is EmployeeName
        // In the mapPartition  one can intialize and shutdown the service gracefully easily.
        // it returns only once when the whole partition is executed , 
        // the result has to held in memory (may cause OOME)
        JavaRDD<Row> returnedRowJavaRDD = rowJavaRDD.mapPartitions
                             (new InsertIntoAerospike(aerospikeHostname,
                                           aerospikePort, namespace, setName, keyName,valueName));

        //MapPartition is just a transformation in Spark hence a action 
        //like write(), collect() is needed to carry out the action.
        // Remeber spark does Lazy evaluation, without a action transformations will not execute.

        DataFrame outputDf = sqlContext.createDataFrame(returnedRowJavaRDD, outPutSchemaStructType);

        // Convert JavaRDD to dataframe and save into parquet file

        return 0;

    // Do remember all the lambda function are instantiated on driver, serialized and sent to driver.
    // No need to initialize the Service(Aerospike , Hbase on driver ) hence making it transient
    // In the call() , for each record insert into Aerospike , this can be converted into batch too
    public static class InsertIntoAerospike implements
                                          FlatMapFunction<Iterator<Row>, Row> {

        // not making it static , as it will not be serialized and sent to executors
        private final String aerospikeHostName;
        private final int aerospikePortNo;
        private final String aerospikeNamespace;
        private final String aerospikeSetName;
        private final String keyColumnName;
        private final String valueColumnName;

        // The Aerospike client is not serializable 
        // and neither there is a need to instantiate on driver
        private transient AerospikeClient client;
        private transient WritePolicy policy;

        public InsertIntoAerospike(String hostName, int portNo, String nameSpace,
                                    String setName, String keyColumnName, String valueColumnName){
            this.aerospikeHostName = hostName;
            this.aerospikePortNo = portNo;
            this.aerospikeNamespace = nameSpace;
            this.aerospikeSetName = setName;
            this.keyColumnName = keyColumnName;
            this.valueColumnName = valueColumnName;


        public Iterable<Row> call(Iterator<Row> rowIterator) throws Exception {

         // rowIterator   points to the whole partition 
        // (all the records in the partition) , not a single record.
        // The partition can be of Map Task or Reduce Task
        // This is where you initialize your service ,
       // MapPartition resemble setup , map , cleanup of MapReduce,
       // Drawback of mapPartition: 
       //it returns after processing the whole chunk not after every row.

            // This will be held in memory till whole partition is evaluated
            List<Row> rowList = new ArrayList<>();

            //This is the place where you initialise your service
            policy = new WritePolicy();
            client = new AerospikeClient(aerospikeHostName, aerospikePortNo);

            // All the code before iterating over rowIterator will be executed only once 
            // as the call() is called only once for each partition

            while (rowIterator.hasNext()) {

                Row row = rowIterator.next();
                // As rows have schema with fieldName and Values being part of the Row
                Key key = new Key(aerospikeNamespace, aerospikeSetName,
                                            (Integer) row.get(row.fieldIndex(keyColumnName)));
                Bin bin = new Bin(valueColumnName, (String)row
                client.put(policy, key, bin);


            //Shutdown the service gracefully as this is called only once for the whole partition

            return rowList;

    public void close() throws IOException {

    public static void main(String[] args) throws Exception {
        ToolRunner.run(new ReadFromHdfsWriteToAerospikeSparkMapPartition(), args);

Key Take Aways:

  • mapPartitons do provide a clean way to intialize and shutdown a service
  • mapPartions needs to keep the output in memory and returns only when the whole partition is executed.


Aerospike Unit Test framework:


Integration Test:


package com.big.data.spark;

import com.aerospike.client.AerospikeClient;
import com.aerospike.client.Key;
import com.aerospike.client.Record;
import com.aerospike.client.policy.WritePolicy;
import com.aerospike.unit.impls.AerospikeRunTimeConfig;
import com.aerospike.unit.impls.AerospikeSingleNodeCluster;
import com.aerospike.unit.utils.AerospikeUtils;
import com.big.data.avro.AvroUtils;
import com.big.data.avro.schema.Employee;
import org.apache.avro.generic.GenericData;
import org.apache.avro.generic.GenericRecord;
import org.apache.commons.io.FileUtils;
import org.apache.hadoop.fs.Path;
import org.junit.After;
import org.junit.Assert;
import org.junit.Before;
import org.junit.Test;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import parquet.avro.AvroParquetReader;
import parquet.avro.AvroParquetWriter;
import parquet.hadoop.ParquetReader;
import parquet.hadoop.ParquetWriter;
import parquet.hadoop.metadata.CompressionCodecName;

import java.io.File;

public class ReadFromHdfsWriteToAerospikeSparkMapPartitionTest {
    private static final Logger LOG = LoggerFactory.getLogger(ReadWriteAvroParquetFilesTest.class);
    private static final String BASEDIR = 
    "/tmp/ReadFromHdfsWriteToASMapPartition/avroparquetInputFile/" +System.currentTimeMillis()+ "/";
    private String input;
    private String output;

    private AerospikeSingleNodeCluster cluster;
    private AerospikeRunTimeConfig runtimConfig;
    private AerospikeClient client;

    //Aerospike unit related Params
    private String memorySize = "64M";
    private String setName = "BIGTABLE";
    private String binName = "BIGBIN";
    private String nameSpace = "RandomNameSpace";

    public void setUp() throws Exception {

        input = BASEDIR + "input/";
        output = BASEDIR + "output/";

        Employee employee = new Employee();

        Employee employee1 = new Employee();

        //Write parquet file2 with GZIP compression
        ParquetWriter<Object> writer = AvroParquetWriter
                                              .builder(new Path(input + "1.gz.parquet"))
                                              .withCompressionCodec (CompressionCodecName.GZIP)

        // Start Aerospike MiniCluster
        // Instatiate the cluster with NameSpaceName , memory Size.
        // One can use the default constructor and retrieve nameSpace,
        //Memory info from cluster.getRunTimeConfiguration();
        cluster = new AerospikeSingleNodeCluster(nameSpace, memorySize);
        // Get the runTime configuration of the cluster
        runtimConfig = cluster.getRunTimeConfiguration();
        client = new AerospikeClient("", runtimConfig.getServicePort());


    public void testSuccess() throws Exception {

    String[] args = new String[]{"-D" + ReadWriteAvroParquetFiles.INPUT_PATH + "=" + input,
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.OUTPUT_PATH + "=" + output,
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.IS_RUN_LOCALLY + "=true",
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.DEFAULT_FS + "=file:///",
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.NUM_PARTITIONS + "=1",
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.AEROSPIKE_NAMESPACE + "=" 
                                                                + runtimConfig.getNameSpaceName(),
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.AEROSPIKE_HOSTNAME + "=",
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.AEROSPIKE_PORT + "=" 
                                                                + runtimConfig.getServicePort(),
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.AEROSPIKE_SETNAME + "=" + setName,
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.KEY_NAME + "=emp_id",
            "-D" + ReadFromHdfsWriteToAerospikeSparkMapPartition.VALUE_NAME + "=emp_name"};


        ParquetReader<GenericRecord> reader = AvroParquetReader.builder(new Path(output)).build();
        //Use .withConf(FS.getConf()) for reading from a diferent HDFS and not local ,
        // by default the fs is local
        GenericData.Record event = null;
        while ((event = (GenericData.Record) reader.read()) != null) {
            Employee outputEvent = AvroUtils.convertByteArraytoAvroPojo
                                          Employee.getClassSchema()), Employee.getClassSchema());
            LOG.info("Data read from Sparkoutput is {}", outputEvent.toString());
            Assert.assertTrue(outputEvent.getEmpId().equals(1) || outputEvent.getEmpId().equals(2));

        //Fetch both the keys from Aerospike
        WritePolicy policy = new WritePolicy();
        Key key1 = new Key(runtimConfig.getNameSpaceName(), setName, 1);
        Record result1 = client.get(policy, key1);
        Assert.assertEquals(result1.getValue("emp_name").toString(), "Maverick1");

        Key key2 = new Key(runtimConfig.getNameSpaceName(), setName, 2);
        Record result2 = client.get(policy, key2);
        Assert.assertEquals(result2.getValue("emp_name").toString(), "Maverick2");


    public void cleanup() throws Exception {
        FileUtils.deleteDirectory(new File(BASEDIR));

        if (cluster != null) {
            // Stop the cluster
        if (client != null) {