Spark tutorial - Understanding and exploring spark core components RDDs

Spark is fast, ease to use data processing engine. Spark is developed as replacement of Hadoop MapReduce and runs over Hadoop. It uses Hadoop Distributed File System of Hadoop. Spark is the open source apache product available since 2014. Spark popularity is growing at rapid speed. Spark doesn't need extra skill set. Knowledge of Core Java and Distributing computing is enough for it.

There are terminology which founds difficults to understand by spark developers. I am going to explain those in my own language.

Spark RDDs 

Spark RDDs is the resilient distributed dataset. RDDs is immutable datasets which can be created from internal source (i.e. Parallelize way) or external source (i.e. Spark streaming, text file, input file format etc). RDD's elements can distributed across spark cluster for processing. RDDs can only be created by reading data from a stable storage such as TCP Streaming, Files or by transformations on existing RDDs.


For example- If as a spark developer you write spark TCP streaming job with interval of 1 seconds. Spark continuously receive data and form a RDD form the data receives in that one seconds.
Look at the image below. RDD consists of elemens E1, E2 .... En which can be json, text, line or any other string. This RDD can be transfom into new RDD but here i am not talking about this.
Elements of RDDs can be processed using foreach  element loop. And then these elements will be distributed across executors in spark clusters.

Spark RDDs

Point (1) -  This explain how RDD is created from the data received on spark tcp socket in 1 seconds. 

Point (2) - This explain that elements of RDD assigned to executors for processing. 

Below is the code of simple spark streaming. Where socket is opend to a IP and port and RDDs are formed every 3s.See clearly where point 1 and 2 lies in the codes.

 SparkConf sparkConf = new SparkConf().setMaster("spark://10.1.0.5:8088").setAppName("App name");  
           JavaStreamingContext ssc = new JavaStreamingContext(sparkConf, new Duration(3000));  
           JavaDStream<String> stream = ssc.socketTextStream("socket IP", 9000, StorageLevels.MEMORY_AND_DISK_SER);  
           stream.foreachRDD(new VoidFunction<JavaRDD<String>>() {  
                private static final long serialVersionUID = 1L;  
                public void call(JavaRDD<String> rdd) throws Exception {  

                     /** Point (1) **/  
                     rdd.foreach(new VoidFunction<String>() {  
                          private static final long serialVersionUID = 1L;  
                          public void call(String s) throws Exception {  

                               /** Point (2) **/  
                               System.out.println(s);  
                          }  
                     });  
                }  
           });  
           ssc.start();  
           ssc.awaitTermination();   

Spark TCP streaming example without Kafka

Spark Streaming is an extension of the core Spark API that enables scalable, high-throughput, fault-tolerant stream processing of live data streams. Data can be ingested from many sources like Kafka, Flume, Kinesis, or TCP sockets. And Finally, processed data can be pushed to filesystems, databases, and live dashboards. On data, you can apply Spark’s machine learning and graph processing algorithms on data streams.

Spark streaming is useful to read data from producer and distribute data over multiple machine in clustor or yarn mode.

Few term related to spark streaming -

RDD stands for resilient data distribution. RDD is created from the data that bring when spark streaming executes in batch interval.

Creating simple example of TCP socket streaming is given below -

 @SuppressWarnings("resource")
 public static void main(String[] args) {
  
  SparkConf sparkConf = new SparkConf().setMaster("spark-master-url").setAppName("xyz")
    .set("spark.executor.memory", "1g").set("spark.cores.max", "5").set("spark.driver.cores", "2")
    .set("spark.driver.memory", "2g");
 
  JavaStreamingContext ssc = new JavaStreamingContext(sparkConf, new Duration(3000));

  JavaDStream<String> JsonReq1 = ssc.socketTextStream("bindIP", bindport, StorageLevels.MEMORY_AND_DISK_SER);
  JavaDStream<String> JsonReq2 = ssc.socketTextStream("bindIP", bindport, StorageLevels.MEMORY_AND_DISK_SER);
  ArrayList<JavaDStream<String>> streamList = new ArrayList<JavaDStream<String>>();
  streamList.add(JsonReq1);
  JavaDStream<String> UnionStream = ssc.union(JsonReq2, streamList);

  UnionStream.foreachRDD(new VoidFunction<JavaRDD<String>>() {

   private static final long serialVersionUID = 1L;

   public void call(JavaRDD<String> rdd) throws Exception {

   
    rdd.foreach(new VoidFunction<String>() {

     private static final long serialVersionUID = 1L;

     public void call(String s) throws Exception {
      System.out.println(s);
     }

    });
   }
  });

  System.out.println(UnionStream.count());
  ssc.start();
  ssc.awaitTermination();
 }

Term like bindIP and bindport will be your specific spark ip/port. To test this application you can create a basic service socket port programm which must listen for clients socket from spark executor.
spark-master-url should be the url of machine where spark master is running . spark master url generally looks like spark://machineip:port