why can i run spark app in eclipse directly withou

问题:

1.My spark(standalone) cluster: spmaster,spslave1,spslave2

2.For my simple spark app which selects some records from mysql.

public static void main(String[] args) {
    SparkConf conf = new SparkConf()
            .setMaster("spark://spmaster:7077")
            .setAppName("SparkApp")
            .set("spark.driver.extraClassPath","/usr/lib/spark-1.6.1-bin-hadoop2.6/lib/mysql-connector-java-5.1.24.jar")  //the driver jar was uploaded to all nodes
            .set("spark.executor.extraClassPath","/usr/lib/spark-1.6.1-bin-hadoop2.6/lib/mysql-connector-java-5.1.24.jar");
    JavaSparkContext sc = new JavaSparkContext(conf);
    SQLContext sqlContext = new SQLContext(sc);
    String url = "jdbc:mysql://192.168.31.43:3306/mytest";
    Map<String, String> options = new HashMap<String, String>();
    options.put("url", url);
    options.put("dbtable", "mytable");
    options.put("user", "root");
    options.put("password", "password");
    DataFrame jdbcDF = sqlContext.read().format("jdbc").options(options)
            .load();
    jdbcDF.registerTempTable("c_picrecord");
    DataFrame sql = sqlContext.sql("select * from mytable limit 10");
    sql.show();  // **show the result on eclipse console**
    sc.close();
}

3.My question : when i right click->run as 'Java Application', it works successfully, and i can find the job on webUI<spark://spmaster:7077>.I don't undersatand how it works , and what is the different between with using spark-submit.sh.

回答1:

When we use spark-submit.sh for submitting application, then spark-submit already created Spark Context (aka Driver) by default.

But when we use Java API (JavaSparkContext) to connect master, then Java application will become Driver. And by using this Driver all application/job will submitted to master.

回答2:

The spark-submit.sh script is just a wrapper around a ${JAVA_HOME}/bin/java execution command. It sets up the environment details and then runs something like:

${JAVA_HOME}/bin/java -Xmx128m -cp "$LAUNCH_CLASSPATH" org.apache.spark.launcher.Main "$@"

When you click on run as 'Java Application' you're also triggering a java execution command, but without all the environment settings done by spark-submit.sh and with the differences mentioned by @Sheel.