1.1.1 Case 2: Spark SQL Development Example
1.1.1.1 Scenario
Applicable Versions
FusionInsight HD V100R002C70, FusionInsight HD V100R002C80
Scenario
Develop a Spark application to perform the following operations on logs about netizens who dwell on online shopping on a weekend.
l Collect statistics on female netizens who dwell on online shopping for over 2 hours on the weekend.
l The first column in the log file records names, the second column records gender, and the third column records the dwell duration in the unit of minute. Three columns are separated by comma (,).
log1.txt: logs collected on Saturday
LiuYang,female,20
YuanJing,male,10
GuoYijun,male,5
CaiXuyu,female,50
Liyuan,male,20
FangBo,female,50
LiuYang,female,20
YuanJing,male,10
GuoYijun,male,50
CaiXuyu,female,50
FangBo,female,60
log2.txt: logs collected on Sunday
LiuYang,female,20
YuanJing,male,10
CaiXuyu,female,50
FangBo,female,50
GuoYijun,male,5
CaiXuyu,female,50
Liyuan,male,20
CaiXuyu,female,50
FangBo,female,50
LiuYang,female,20
YuanJing,male,10
FangBo,female,50
GuoYijun,male,50
CaiXuyu,female,50
FangBo,female,60
Data Planning
Save the original log files in the HDFS.
1. Create two text files input_data1.txt and input_data2.txt on the local host, copy the content in log1.txt to input_data1.txt, and copy the content in log2.txt to input_data2.txt.
2. Create the /tmp/input folder in the HDFS, and run the following commands to upload input_data1.txt and input_data2.txt to the /tmp/input directory:
a. On the HDFS client, run the following commands for authentication:
cd/opt/hadoopclient
kinit< service user for authentication >
b. On the HDFS client of the Linux OS, run the hadoop fs -mkdir/tmp/input command (the hdfs dfs command has the same function) to create a directory.
c. On the HDFS client of the Linux OS, run the hadoop fs -putinput_data1.txt/tmp/input and hadoop fs -putinput_data2.txt/tmp/input commands to upload data files.
1.1.1.2 Development Guidelines
Collect statistics on female netizens who dwell on online shopping for over 2 hours on the weekend.
To achieve the objective, the process is as follows:
l Create a table and import the log files into the table.
l Filter data information of the time that female netizens spend online.
l Summarize the total time that each female netizen spends online.
l Filter the information of female netizens who spend over 2 hours online.
1.1.1.3 Sample Code Description
1.1.1.3.1 Java Code Example
Function
Collect statistics on female netizens who dwell on online shopping for over 2 hours on the weekend.
Sample Code
The following code snippets are used as an example. For complete codes, see the com.huawei.bigdata.spark.examples.FemaleInfoCollection class.
SparkConf conf = new SparkConf().setAppName("CollectFemaleInfo");
JavaSparkContext jsc = new JavaSparkContext(conf);
SQLContext sqlContext = new org.apache.spark.sql.SQLContext(jsc);
// Convert RDD into DataFrame using the implicit conversion.
JavaRDD<FemaleInfo> femaleInfoJavaRDD = jsc.textFile(args[0]).map(
new Function<String, FemaleInfo>() {
@Override
public FemaleInfo call(String line) throws Exception {
String[] parts = line.split(",");
FemaleInfo femaleInfo = new FemaleInfo();
femaleInfo.setName(parts[0]);
femaleInfo.setGender(parts[1]);
femaleInfo.setStayTime(Integer.parseInt(parts[2].trim()));
return femaleInfo;
}
});
// Register a table.
DataFrame schemaFemaleInfo = sqlContext.createDataFrame(femaleInfoJavaRDD,FemaleInfo.class);
schemaFemaleInfo.registerTempTable("FemaleInfoTable");
// Execute the SQL query.
DataFrame femaleTimeInfo = sqlContext.sql("select * from " +
"(select name,sum(stayTime) as totalStayTime from FemaleInfoTable " +
"where gender = 'female' group by name )" +
" tmp where totalStayTime >120");
// The command output is displayed.
List<String> result = femaleTimeInfo.javaRDD().map(new Function<Row, String>() {
public String call(Row row) {
return row.getString(0) + "," + row.getLong(1);
}
}).collect();
System.out.println(result);
jsc.stop();
The preceding is a simple example. For details about other sparkSQL features, see the following link: http://spark.apache.org/docs/latest/sql-programming-guide.html#running-sql-queries-programmatically.
1.1.1.3.2 Scala Code Example
Function
Collect statistics on female netizens who dwell on online shopping for over 2 hours on the weekend.
Sample Code
The following code snippets are used as an example. For complete code, see the com.huawei.bigdata.spark.examples.FemaleInfoCollection class.
object CollectFemaleInfo {
// The table structure is used to map text data to df.
case class FemaleInfo(name: String, gender: String, stayTime: Int)
def main(args: Array[String]) {
// Set the Spark application name.
val sparkConf = new SparkConf().setAppName("FemaleInfo")
val sc = new SparkContext(sparkConf)
val sqlContext = new org.apache.spark.sql.SQLContext(sc)
import sqlContext.implicits._
<}// Convert the RDD into the DataFrame using the implicit conversion, and then register a table.
sc.textFile(args(0)).map(_.split(","))
.map(p => FemaleInfo(p(0), p(1), p(2).trim.toInt))
.toDF.registerTempTable("FemaleInfoTable")
// Use SQL statements to filter female Internet access time data and aggregate the same name online.
val femaleTimeInfo = sqlContext.sql("select name,sum(stayTime) as stayTime from FemaleInfoTable where
gender = 'female' group by name")
<}// Screen the information about female netizens who spend over two hours and generate the information.
val c = femaleTimeInfo.filter("stayTime >= 120").collect().foreach(println)
sc.stop()
}
}
The preceding is a simple example. For details about other sparkSQL features, see the following link: http://spark.apache.org/docs/latest/sql-programming-guide.html#running-sql-queries-programmatically.
1.1.1.4 Obtaining Sample Code
Using the FusionInsight Client
Obtain the sample project in the sampleCode directory in the Spark directory in the FusionInsight_Services_ClientConfig file extracted from the client.
Security mode: SparkSQLJavaExample and SparkSQLScalaExample in the spark-examples-security directory
Non-security mode: SparkSQLJavaExample and SparkSQLScalaExample in the spark-examples-normal directory
Using the Maven Project
log in to Huawei DevClod (https://codehub-cn-south-1.devcloud.huaweicloud.com/codehub/7076065/home) to download code udner to local PC.
Security mode:
components/spark/spark-examples-security/SparkJavaExample
components/spark/spark-examples-security/SparkScalaExample
Non-security mode:
components/spark/spark-examples-normal/SparkJavaExample
components/spark/spark-examples-normal/SparkScalaExample
1.1.1.5 Application Commissioning
1.1.1.5.1 Compiling and Running the Application
Scenario
After the program code is developed, you can upload the code to the Linux client for running. The running procedures of applications developed in Scala or Java are the same.
l The Spark application can run only in the Linux environment but not in the Windows environment.
l The Spark application developed in Python does not need to build Artifacts as a jar. You just need to copy the sample projects to the compiler.
l It is needed to ensure that the version of Python installed on the worker and driver is consistent, otherwise the following error will be reported: "Python in worker has different version %s than that in driver %s."
Procedure
Step 1 In the IntelliJ IDEA, configure the Artifacts information about the project before the jar is created.
1. On the main page of the IDEA, choose File > Project Structures... to enter the Project Structure page.
2. On the Project Structure page, select Artifacts, click + and choose Jar > From modules with dependencies....
Figure 1-1 Add the Artifacts
3. Select the corresponding module. The module corresponding to the Java sample projects is CollectFemaleInfo. Click OK.
Figure 1-2 Create Jar from Modules
4. Configure the name, type and output directory of the Jar based on the actual condition.
Figure 1-3 Configure the basic information
5. Right-click CollectFemaleInfo, choose Put into Output Root, and click Apply.
Figure 1-4 Put into Output Root
6. Click OK to complete the configuration.
Step 2 Create the jar.
1. On the main page of the IDEA, choose Build > Build Artifacts....
Figure 1-5 Build Artifacts
2. On the displayed menu, choose CollectFemaleInfo > Build to create a jar.
Figure 1-6 Build
3. If the following information is displayed in the event log, the jar is created successfully. You can obtain the jar from the directory configured in Step 1.4.
21:25:43 Compilation completed successfully in 36 sec
Step 3 Copy the jar created in 2 to the Spark running environment (Spark client), such as /opt/hadoopclient/Spark to run the Spark application.
When a Spark task is running, it is prohibited to restart the HDFS service or restart all DataNode instances. Otherwise, the Spark task may fail, resulting in JobHistory data loss.
l Run the sample projects of Spark Core (including Scala and Java).
Access the Spark client directory and implement the bin/spark-submit script to run the codes.
<inputPath> indicates the input directory in the HDFS.
bin/spark-submit --classcom.huawei.bigdata.spark.examples.FemaleInfoCollection --master yarn-client/opt/female/FemaleInfoCollection.jar <inputPath>
l Run the sample projects of Spark SQL (Java and Scala).
Access the Spark client directory and implement the bin/spark-submit script to run the codes.
<inputPath> indicates the input directory in the HDFS.
bin/spark-submit --classcom.huawei.bigdata.spark.examples.FemaleInfoCollection --master yarn-client/opt/female/FemaleInfoCollection.jar <inputPath>
l Run the sample projects of Spark Streaming (Java and Scala).
Access the Spark client directory and implement the bin/spark-submit script to run the codes.
The location of Spark Streaming Kafka dependency package on the client is different from the location of other dependency packages. For example, the path to the Spark Streaming Kafka dependency package is $SPARK_HOME/lib/streamingClient, whereas the path to other dependency packages is $SPARK_HOME/lib. When running an application, you must add the configuration option to the spark-submit command to specify the path of Spark Streaming Kafka dependency package. The following is an example path:
--jars $SPARK_HOME/lib/streamingClient/kafka-clients-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/kafka_2.10-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/spark-streaming-kafka_2.10-1.5.1.jar
Example codes of the Spark Streaming Write To Print is as follows:
bin/spark-submit --master yarn-client--jars $SPARK_HOME/lib/streamingClient/kafka-clients-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/kafka_2.10-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/spark-streaming-kafka_2.10-1.5.1.jar --classcom.huawei.bigdata.spark.examples.FemaleInfoCollectionPrint /opt/female/FemaleInfoCollectionPrint.jar <checkPointDir> <batchTime> <topics> <brokers>
Example codes of the Spark Streaming Write To Kafka is as follows:
bin/spark-submit --master yarn-client--jars $SPARK_HOME/lib/streamingClient/kafka-clients-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/kafka_2.10-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/spark-streaming-kafka_2.10-1.5.1.jar --classcom.huawei.bigdata.spark.examples.FemaleInfoCollectionKafka /opt/female/FemaleInfoCollectionKafka.jar <checkPointDir> <batchTime> <windowTime> <topics> <brokers>
l Run the sample projects of Accessing the Spark SQL Through JDBC (Java and Scala).
Access the Spark client directory and implement the java -cp command to run the codes.
java -cp$SPARK_HOME/lib/*:$SPARK_HOME/conf:/opt/female/ThriftServerQueriesTest.jar com.huawei.bigdata.spark.examples.ThriftServerQueriesTest $SPARK_HOME/conf/hive-site.xml $SPARK_HOME/conf/spark-defaults.conf
In the preceding command line, you can choose the minimal runtime dependency package based on the sample projects. For details of the runtime dependency packages, see References.
l Run the Spark on HBase sample application(Java and Scala).
a. Verify that the configuration options in the Spark client configuration file spark-defaults.conf are correctly configured.
When running the Spark on HBase sample application, set the configuration option spark.hbase.obtainToken.enabled in the Spark client configuration file spark-defaults.conf to true (The default value is false. Changing the value to true does not affect existing services. If you want to uninstall the HBase service, change the value back to false first. Set the configuration option spark.inputFormat.cache.enabled to false.
Table 1-1 Parameters
|
Parameter |
Description |
Default Value |
|
spark.hbase.obtainToken.enabled |
Indicates whether to enable the function of obtaining the HBase token. |
false |
|
spark.inputFormat.cache.enabled |
Indicates whether to cache the InputFormat that maps to HadoopRDD. If the parameter is set to true, the tasks of the same Executor use the same InputFormat object. In this case, the InputFormat must be thread-safe. If caching the InputFormat is not required, set the parameter to false. |
true |
b. Access the Spark client directory and implement the bin/spark-submit script to run the code.
Run sample applications in the sequence: TableCreation > TableInputData > TableOutputData.
When the TableInputData sample application is running, <inputPath> needs to be specified. <inputPath>indicates the input path in the HDFS.
bin/spark-submit --classcom.huawei.bigdata.spark.examples.TableInputData --master yarn-client/opt/female/TableInputData.jar <inputPath>
l Run the Spark Hbase to HBase sample application(Scala and Java).
Access the Spark client directory and implement the bin/spark-submit script to run the code.
bin/spark-submit --classcom.huawei.bigdata.spark.examples.SparkHbasetoHbase --master yarn-client/opt/female/FemaleInfoCollection.jar
l Run the Spark Hive to HBase sample application(Scala and Java).
Access the Spark client directory and implement the bin/spark-submit script to run the code.
bin/spark-submit --classcom.huawei.bigdata.spark.examples.SparkHivetoHbase --master yarn-client/opt/female/FemaleInfoCollection.jar
l Run the Spark Streaming Kafka to HBase sample application(Scala and Java).
Access the Spark client directory and implement the bin/spark-submit script to run the code.
When the sample application is running, specify the <checkPointDir><topic><brokerList>. <checkPointDir> indicates the directory where the application result is backed up, <topic> indicates the topic that is read from Kafka, <brokerList> indicates the IP address of the Kafka server.
On the client, the directory of Spark Streaming Kafka dependency package is different from the directory of other dependency packages. For example, the directory of another dependency package is $SPARK_HOME/lib and the directory of a Spark Streaming Kafka dependency package is $SPARK_HOME/lib/streamingClient. Therefore, when running the application, add the configuration option in the spark-submit command to specify the directory for the Spark Streaming Kafka dependency package, for example, --jars $SPARK_HOME/lib/streamingClient/kafka-clients-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/kafka_2.10-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/spark-streaming-kafka_2.10-1.5.1.jar.
Example code of Spark Streaming To HBase
bin/spark-submit --master yarn-client --jars$SPARK_HOME/lib/streamingClient/kafka-clients-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/kafka_2.10-0.8.2.1.jar,$SPARK_HOME/lib/streamingClient/spark-streaming-kafka_2.10-1.5.1.jar --classcom.huawei.bigdata.spark.examples.streaming.SparkOnStreamingToHbase /opt/female/FemaleInfoCollectionPrint.jar <checkPointDir> <topic> <brokerList>
l Submit the application developed in Python.
Access the Spark client directory and implement the bin/spark-submit script to run the codes.
<inputPath> indicates the input directory in the HDFS.
Because the sample code does not contain the authentication information, specify the authentication information by configuring the spark.yarn.keytab and spark.yarn.principle when the application is run.
bin/spark-submit --master yarn-client --conf spark.yarn.keytab=/opt/FIclient/user.keytab --conf spark.yarn.principal=sparkuser/opt/female/SparkPythonExample/collectFemaleInfo.py <inputPath>
----End
References
The runtime dependency packages for the sample projects of Accessing the Spark SQL Through JDBC (Java and Scala) are as follows:
l The sample projects of Accessing the Spark SQL Through JDBC (Scala):
− avro-1.7.7.jar
− commons-collections-3.2.2.jar
− commons-configuration-1.6.jar
− commons-io-2.4.jar
− commons-lang-2.6.jar
− commons-logging-1.1.3.jar
− guava-12.0.1.jar
− hadoop-auth-2.7.2.jar
− hadoop-common-2.7.2.jar
− hadoop-mapreduce-client-core-2.7.2.jar
− hive-exec-1.2.1.spark.jar
− hive-jdbc-1.2.1.spark.jar
− hive-metastore-1.2.1.spark.jar
− hive-service-1.2.1.spark.jar
− httpclient-4.5.2.jar
− httpcore-4.4.4.jar
− libthrift-0.9.3.jar
− log4j-1.2.17.jar
− slf4j-api-1.7.10.jar
− zookeeper-3.5.1.jar
− scala-library-2.10.4.jar
l The sample projects of Accessing the Spark SQL Through JDBC (Java):
− commons-collections-3.2.2.jar
− commons-configuration-1.6.jar
− commons-io-2.4.jar
− commons-lang-2.6.jar
− commons-logging-1.1.3.jar
− guava-2.0.1.jar
− hadoop-auth-2.7.2.jar
− hadoop-common-2.7.2.jar
− hadoop-mapreduce-client-core-2.7.2.jar
− hive-exec-1.2.1.spark.jar
− hive-jdbc-1.2.1.spark.jar
− hive-metastore-1.2.1.spark.jar
− hive-service-1.2.1.spark.jar
− httpclient-4.5.2.jar
− httpcore-4.4.4.jar
− libthrift-0.9.3.jar
− log4j-1.2.17.jar
− slf4j-api-1.7.10.jar
− zookeeper-3.5.1.jar
1.1.1.5.2 Checking the Commissioning Result
Scenario
After a Spark application is run, you can check the running result through one of the following methods:
l Viewing the command output.
l Logging in to the Spark WebUI.
l Viewing Spark logs.
Procedure
l Check the operating result data of the Spark application.
The data storage directory and format are specified by users in the Spark application. You can obtain the data in the specified file.
l Check the status of the Spark application.
The Spark contains the following two Web UIs:
− The Spark UI displays the status of applications being executed.
The Spark UI contains the Spark Jobs, Spark Stages, Storage, Environment, and Executors parts. Besides these parts, Spark Streaming is displayed for the Streaming application.
Access to the interface: On the Web UI of the YARN, find the corresponding Spark application, and click the final column ApplicationMaster of the application information to access the Spark UI.
− The History Server UI displays the status of all Spark applications.
The History Server UI displays information such as the application ID, application name, start time, end time, execution time, and user to whom the application belongs. After the application ID is clicked, the Spark UI of the application is displayed.
l View Spark logs to learn application running conditions.
The logs of Spark offers immediate visibility into application running conditions. You can adjust application programs based on the logs. Log related information can be referenced to Spark in the Log Description in the Administrator Guide.
