Streaming MQTT Spark Data

This topic walks you through using MQTT Spark streaming with Splice Machine. MQTT is a lightweight, publish-subscribe messaging protocol designed for connecting remotely when a small footprint is required. MQTT is frequently used for data collection with the Internet of Things (IoT).

The example code in this tutorial uses Mosquitto, which is an open source message broker that implements the MQTT. This tutorial uses a cluster managed by MapR; if you’re using different platform management software, you’ll need to make a few adjustments in how the code is deployed on your cluster.

All of the code used in this tutorial is available in our GitHub community repository.

You can complete this tutorial by watching a short video or by following the written directions below.

Watch the Video

The following video shows you how to:

  • put messages on an MQTT queue
  • consume those messages using Spark streaming
  • save those messages to Splice Machine with a virtual table (VTI)

Written Walk Through

This section walks you through the same sequence of steps as the video, in these sections:

Deploying the Tutorial Code

Follow these steps to deploy the tutorial code:

  1. Download the code from our GitHub community repository.

    Pull the code from our git repository:
  2. Compile and package the code:

    mvn clean compile package
  3. Copy three JAR files to each server:

    Copy these three files:


    to this directory on each server:

  4. Restart Hbase

  5. Create the target table in splice machine:

    Run this script to create the table:

  6. Start Mosquitto:

    sudo su /usr/sbin/mosquitto -d -c /etc/mosquitto/mosquitto.conf > /var/log/mosquitto.log 2>&1
  7. Start the Spark streaming script:

    sudo -su mapr ./ tcp://srv61:1883 /testing 10

    The first parameter (tcp://srv61:1883) is the MQTT broker, the second (/testing) is the topic name, and the third (10) is the number of seconds each stream should run.

  8. Start putting messages on the queue:

    Here’s a java program that is set up to put messages on the queue:

    java -cp /opt/splice/default/lib/splice-tutorial-mqtt-2.0-SNAPSHOT.jar:/opt/splice/default/lib/org.eclipse.paho.client.mqttv3-1.1.0.jar com.splicemachine.tutorials.sparkstreaming.mqtt.MQTTPublisher tcp://localhost:1883 /testing 1000 R1

    The first parameter (tcp://localhost:1883) is the MQTT broker, the second (/testing) is the topic name, the third (1000) is the number of iterations to execute, and the fourth parameter (R1) is a prefix for this run.

    The source code for this utility program is in a different GitHub project than the rest of this code. You’ll find it in the  tutorial-kafka-producer Github project.

About the Sample Code

This section describes the main class methods used in this MQTT example code; here’s a summary of the classes:

Java Class Description
MQTTPublisher Puts csv messages on an MQTT queue.
SparkStreamingMQTT The Spark streaming job that reads messages from the MQTT queue.
SaveRDD Inserts the data into Splice Machine using the RFIDMessageVTI class.
RFIDMessageVTI A virtual table interface for parsing an RFIDMessage.
RFIDMessage Java object (a POJO) for converting from a csv string to an object to a database entry.


This class puts CSV messages on an MQTT queue. The function of most interest in is DoDemo, which controls our sample program:

public void doDemo() {
   try {
        long startTime = System.currentTimeMillis();
        client = new MqttClient(broker, clientId);
        MqttMessage message = new MqttMessage();
        for (int i=0; inumMessages; i++) {
            // Build a csv string
            message.setPayload( prefix + "Asset" + i ", Location" + i + "," + new Timestamp((new Date()).getTime())).getBytes());
            client.publish(topicName, message);
            if (i % 1000 == 0) {
                System.out.println("records:" + i + " duration=" + (System.currentTimeMillis() - startTime));
                startTime = System.currentTimeMillis();
    } catch (MqttException e) {

DoDemo does a little initialization, then starts putting messages out on the queue. Our sample program is set up to loop until it creates numMessages messages; after every 1000 messages, it displays a status message that helps us determine how much time is going to put messages on the queue, and how much to take them off the queue.

DoDemo builds a csv record (line) for each message, setting an asset ID, a location ID, and a timestamp in the payload of the message. It them publishes that message to the topic topicName.


Once the messages are on the queue, our SparkStreamingMQTT class object reads them from the queue and inserts them into our database. The main method in this class is processMQTT:

public void processMQTT(final String broker, final String topic, final int numSeconds) {"************ SparkStreamingMQTTOutside.processMQTT start");

    // Create the spark application and set the name to MQTT
    SparkConf sparkConf = new SparkConf().setAppName("MQTT");

    // Create the spark streaming context with a 'numSeconds' second batch size
    jssc = new JavaStreamingContext(sparkConf, Durations.seconds(numSeconds));
    jssc.checkpoint(checkpointDirectory);"************ SparkStreamingMQTTOutside.processMQTT about to read the MQTTUtils.createStream");
    //2. MQTTUtils to collect MQTT messages
    JavaReceiverInputDStreamString> messages = MQTTUtils.createStream(jssc, broker, topic);"************ SparkStreamingMQTTOutside.processMQTT about to do foreachRDD");
    //process the messages on the queue and save them to the database
    messages.foreachRDD(new SaveRDD());"************ SparkStreamingMQTTOutside.processMQTT prior to context.strt");
    // Start the context

The processMQTT method takes three parameters:


The URL of the MQTT broker.


The MQTT topic name.


The number of seconds at which streaming data will be divided into batches.

The processMQTT method processes the messages on the queue and saves them by calling the SaveMDD class.


The SaveRDD class is an example of a Spark streaming function that uses our virtual table interface (VTI) to insert data into your Splice Machine database. This function checks for messages in the stream, and if there any, it creates a connection your database and uses a prepared statement to insert the messages into the database.

 * This is an example of spark streaming function that
 * inserts data into Splice Machine using a VTI.
 * @author Erin Driggers

public class SaveRDD implements FunctionJavaRDDString>, Void>, Externalizable {

    private static final Logger LOG = Logger.getLogger(SaveRDD.class);

    public Void call(JavaRDDString> rddRFIDMessages) throws Exception {
        LOG.debug("About to read results:");
        if (rddRFIDMessages != null '& rddRFIDMessages.count() > 0) {
            LOG.debug("Data to process:");
            //Convert to list
            ListString> rfidMessages = rddRFIDMessages.collect();
            int numRcds = rfidMessages.size();

            if (numRcds > 0) {
                try {
                    Connection con = DriverManager.getConnection("jdbc:splice://localhost:1527/splicedb;user=splice;password=admin");

                    //Syntax for using a class instance in a VTI, this could also be a table function
                    String vtiStatement = "INSERT INTO IOT.RFID "
                            + "select s.* from new com.splicemachine.tutorials.sparkstreaming.mqtt.RFIDMessageVTI(?) s ("
                            + RFIDMessage.getTableDefinition() + ")";
                    PreparedStatement ps = con.prepareStatement(vtiStatement);
                    ps.setObject(1, rfidMessages);
                } catch (Exception e) {
                    //It is important to catch the exceptions as log messages because it is difficult
                    //to trace what is happening otherwise
                    LOG.error("Exception saving MQTT records to the database" + e.getMessage(), e);
                } finally {
          "Complete insert into IOT.RFID");
        return null;

The heart of this function is the statement that creates the prepared statement, using a VTI class instance:

String vtiStatement = "INSERT INTO IOT.RFID "     + "select s.* from new     com.splicemachine.tutorials.sparkstreaming.mqtt.RFIDMessageVTI(?) s ("
    + RFIDMessage.getTableDefinition() + ")";
PreparedStatement ps = con.prepareStatement(vtiStatement);

Note that the statement references both our RFIDMessage and RFIDMessageVTI classes, which are described below.


The RFIDMessageVTI class implements an example of a virtual table interface that reads in a list of strings that are in CSV format, converts that into an RFIDMessage object, and returns the resultant list in a format that is compatible with Splice Machine.

This class features an override of the getDataSet method, which loops through each CSV record from the input stream and converts it into an RFIDMessage object that is added onto a list of message items:

public DataSetLocatedRow> getDataSet(SpliceOperation op, DataSetProcessor dsp, ExecRow execRow) throws StandardException {
    operationContext = dsp.createOperationContext(op);

    //Create an arraylist to store the key / value pairs
    ArrayListLocatedRow> items = new ArrayListLocatedRow>();

    try {

        int numRcds = this.records == null ? 0 : this.records.size();

        if (numRcds > 0) {

  "Records to process:" + numRcds);
            //Loop through each record convert to a SensorObject
            //and then set the values
            for (String csvString : records) {
                CsvBeanReader beanReader = new CsvBeanReader(new StringReader(csvString), CsvPreference.STANDARD_PREFERENCE);
                RFIDMessage msg =, header, processors);
                items.add(new LocatedRow(msg.getRow()));
    } catch (Exception e) {
        LOG.error("Exception processing RFIDMessageVTI", e);
    } finally {
    return new ControlDataSet>(items);

For more information about using our virtual table interface, see Using the Splice Machine Virtual Table Interface.


The RFIDMessage class creates a simple Java object (a POJO) that represents an RFID message; we use this to convert an incoming CSV-formatted message into an object. This class includes getters and setters for each of the object properties, plus the getTableDefinition and getRow methods:

 * Used by the VTI to build a Splice Machine compatible resultset
 * @return
 * @throws SQLException
 * @throws StandardException
public ValueRow getRow() throws SQLException, StandardException {
    ValueRow valueRow = new ValueRow(5);
    valueRow.setColumn(1, new SQLVarchar(this.getAssetNumber()));
    valueRow.setColumn(2, new SQLVarchar(this.getAssetDescription()));
    valueRow.setColumn(3, new SQLTimestamp(this.getRecordedTime()));
    valueRow.setColumn(4, new SQLVarchar(this.getAssetType()));
    valueRow.setColumn(5, new SQLVarchar(this.getAssetLocation()));
    return valueRow;

 * Table definition to use when using a VTI that is an instance of a class
 * @return
public static String getTableDefinition() {
    return "ASSET_NUMBER varchar(50), "
    + "ASSET_DESCRIPTION varchar(100), "
    + "ASSET_TYPE VARCHAR(50), "

The getTableDefinition method is a string description of the table into which you’re inserting records; this pretty much replicates the specification you would use in an SQL CREATE_TABLE statement.

The getRow method creates a data row with the appropriate number of columns, uses property getters to set the value of each column, and returns the row as a resultset that is compatible with Splice Machine.

About the Sample Code Scripts

These are also two scripts that we use with this tutorial:

Class Description
/ddl/create-tables.sql A simple SQL script that you can use to have Splice Machine create the table into which RFID messages are stored.
/scripts/ Starts the Spark streaming job.