Using the mapPartition API
Overview
You can use the HBaseContext method to perform operations on HBase in Spark applications and use the mapPartition API to traverse HBase tables in parallel.
Preparing Data
Use HBase tables created in Using the foreachPartition API.
Development Guidelines
- Construct RDDs corresponding to rowkey in HBase tables to be traversed.
- Use the mapPartition API to traverse the data corresponding to the rowkey and perform simple operations.
Preparations
For clusters with the security mode enabled, the Spark Core sample code needs to read two files (user.keytab and krb5.conf). The user.keytab and krb5.conf files are authentication files in the security mode. Download the authentication credentials of the user principal on the FusionInsight Manager page. The user in the sample code is super, change the value to the prepared development user name.
Packaging the Project
- Use the Maven tool provided by IDEA to package the project and generate the JAR file. For details, see Commissioning a Spark Application in a Linux Environment.
- Upload the JAR package to any directory (for example, $SPARK_HOME) on the server where the Spark client is located.
- Upload the user.keytab and krb5.conf files to the server where the client is installed (The file upload path must be the same as the path of the generated JAR file).
To run the Spark on HBase sample project, set spark.yarn.security.credentials.hbase.enabled (false by default) in the spark-defaults.conf file on the Spark client to true. Changing the spark.yarn.security.credentials.hbase.enabled value does not affect existing services. (To uninstall the HBase service, you need to change the value of this parameter back to false.) Set configuration item spark.inputFormat.cache.enabled to false.
Submitting Commands
Assume that the JAR package name is spark-hbaseContext-test-1.0.jar that is stored in the $SPARK_HOME directory on the client. The following commands are executed in the $SPARK_HOME directory, and Java is displayed before the class name of the Java interface. For details, see the sample code.
- yarn-client mode:
Java/Scala version (The class name must be the same as the actual code. The following is only an example.)
bin/spark-submit --master yarn --deploy-mode client --class com.huawei.bigdata.spark.examples.hbasecontext.JavaHBaseMapPartitionExample SparkOnHbaseJavaExample.jar table2
Python version. (The file name must be the same as the actual one. The following is only an example.) Assume that the package name of the corresponding Java code is SparkOnHbaseJavaExample.jar and the package is saved to the current directory.
bin/spark-submit --master yarn --deploy-mode client --jars SparkOnHbaseJavaExample.jar HBaseMapPartitionExample.py table2
- yarn-cluster mode:
Java/Scala version (The class name must be the same as the actual code. The following is only an example.)
bin/spark-submit --master yarn --deploy-mode cluster --class com.huawei.bigdata.spark.examples.hbasecontext.JavaHBaseMapPartitionExample --files /opt/user.keytab,/opt/krb5.conf SparkOnHbaseJavaExample.jar table2
Python version. (The file name must be the same as the actual one. The following is only an example.) Assume that the package name of the corresponding Java code is SparkOnHbaseJavaExample.jar and the package is saved to the current directory.
bin/spark-submit --master yarn --deploy-mode cluster --files /opt/user.keytab,/opt/krb5.conf --jars SparkOnHbaseJavaExample.jar HBaseMapPartitionExample.py table2
Java Sample Code
The following code snippet is only for demonstration. For details about the code, see the JavaHBaseMapPartitionExample file in SparkOnHbaseJavaExample.
public static void main(String args[]) throws IOException {
if(args.length <1){
System.out.println("JavaHBaseMapPartitionExample {tableName} is missing an argument");
return;
}
LoginUtil.loginWithUserKeytab();
final String tableName = args[0];
SparkConf sparkConf = new SparkConf().setAppName("HBaseMapPartitionExample " + tableName);
JavaSparkContext jsc = new JavaSparkContext(sparkConf);
try{
List<byte []> list = new ArrayList();
list.add(Bytes.toBytes("1"));
list.add(Bytes.toBytes("2"));
list.add(Bytes.toBytes("3"));
list.add(Bytes.toBytes("4"));
list.add(Bytes.toBytes("5"));
JavaRDD<byte []> rdd = jsc.parallelize(list);
Configuration hbaseconf = HBaseConfiguration.create();
JavaHBaseContext hbaseContext = new JavaHBaseContext(jsc, hbaseconf);
JavaRDD getrdd = hbaseContext.mapPartitions(rdd, new FlatMapFunction<Tuple2<Iterator<byte[]>,Connection>, Object>() {
public Iterator call(Tuple2<Iterator<byte[]>, Connection> t)
throws Exception {
Table table = t._2.getTable(TableName.valueOf(tableName));
//go through rdd
List<String> list = new ArrayList<String>();
while(t._1.hasNext()){
byte[] bytes = t._1.next();
Result result = table.get(new Get(bytes));
Iterator<Cell> it = result.listCells().iterator();
StringBuilder sb = new StringBuilder();
sb.append(Bytes.toString(result.getRow()) + ":");
while(it.hasNext()){
Cell cell = it.next();
String column = Bytes.toString(cell.getQualifierArray());
if(column.equals("counter")){
sb.append("(" + column + "," + Bytes.toLong(cell.getValueArray()) + ")");
} else {
sb.append("(" + column + "," + Bytes.toString(cell.getValueArray()) + ")");
}
}
list.add(sb.toString());
}
return list.iterator();
}
});
List<byte[]> resultList = getrdd.collect();
if(null == resultList || 0 == resultList.size()){
System.out.println("Nothing matches!");
}else{
for(int i =0; i< resultList.size(); i++){
System.out.println(resultList.get(i));
}
}
} finally {
jsc.stop();
}
}
Scala Sample Code
The following code snippet is only for demonstration. For details about the code, see the HBaseMapPartitionExample file in SparkOnHbaseScalaExample.
def main(args: Array[String]) {
if (args.length < 1) {
println("HBaseMapPartitionExample {tableName} is missing an argument")
return
}
LoginUtil.loginWithUserKeytab()
val tableName = args(0)
val sparkConf = new SparkConf().setAppName("HBaseMapPartitionExample " + tableName)
val sc = new SparkContext(sparkConf)
try {
//[(Array[Byte])]
val rdd = sc.parallelize(Array(
Bytes.toBytes("1"),
Bytes.toBytes("2"),
Bytes.toBytes("3"),
Bytes.toBytes("4"),
Bytes.toBytes("5")))
val conf = HBaseConfiguration.create()
val hbaseContext = new HBaseContext(sc, conf)
val b = new StringBuilder
val getRdd = rdd.hbaseMapPartitions[String](hbaseContext, (it, connection) => {
val table = connection.getTable(TableName.valueOf(tableName))
it.map{r =>
//batching would be faster. This is just an example
val result = table.get(new Get(r))
val it = result.listCells().iterator()
b.append(Bytes.toString(result.getRow) + ":")
while (it.hasNext) {
val cell = it.next()
val q = Bytes.toString(cell.getQualifierArray)
if (q.equals("counter")) {
b.append("(" + q + "," + Bytes.toLong(cell.getValueArray) + ")")
} else {
b.append("(" + q + "," + Bytes.toString(cell.getValueArray) + ")")
}
}
b.toString()
}
})
getRdd.collect().foreach(v => println(v))
} finally {
sc.stop()
}
}
Python Sample Code
The following code snippet is only for demonstration. For details about the code, see the HBaseMapPartitionExample file in SparkOnHbasePythonExample.
# -*- coding:utf-8 -*-
"""
[Note]
1. PySpark does not provide HBase APIs. In this example, Python is used to invoke Java code to implement required operations.
2. If yarn-client is used, ensure that the spark.yarn.security.credentials.hbase.enabledparameter in the spark-defaults.conffile under Spark2x/spark/conf/ is set to true on the Spark2x client.
Set spark.yarn.security.credentials.hbase.enabled to true.
"""
from py4j.java_gateway import java_import
from pyspark.sql import SparkSession
# Create a SparkSession instance.
spark = SparkSession\
.builder\
.appName("JavaHBaseMapPartitionExample")\
.getOrCreate()
# Import the required class to sc._jvm.
java_import(spark._jvm, 'com.huawei.bigdata.spark.examples.hbasecontext.JavaHBaseMapPartitionExample')
# Create a class instance, invoke the method, and transfer the sc._jsc parameter.
spark._jvm.JavaHBaseMapPartitionExample().execute(spark._jsc, sys.argv)
# Stop SparkSession.
spark.stop()
Feedback
Was this page helpful?
Provide feedbackThank you very much for your feedback. We will continue working to improve the documentation.See the reply and handling status in My Cloud VOC.
For any further questions, feel free to contact us through the chatbot.
Chatbot
