A comprehensive wiki with theory summaries and code snippets is available at this link. It can be used as a reference guide at the exam.

bigdata

Repository that contains all the projects developed during the laboratories of the "Big Data: Architectures and Data Analytics" course of the Politecnico di Torino. In the wiki section you can find summaries of the lectures, cheatsheets and code samples.

How to create a new project in IntelliJ starting from the sources

create a new project from source, import the libraries trough the "project structure" menu: the libraries are in the labs/lib folder

How to submit a MapReduce job to the bigdata@polito Hadoop cluster

• Compile the MapReduce Application obtaining the jar file.

In Intellij IDEA, go to File->Project Structure-> Artifacts -> generate artifacts, select as main class the Driver.

• scp the jar file to the gateway using: scp <idea-project-folder>/out/artifacts/main_jar/main.jar s241915@bigdatalab.polito.it:/home/bigdata-01QYD/s241915/

• use HDFS commands to copy the input files into the cluster file system, or connect to: https://bigdatalab.polito.it:8080 to use the web interface (Hue).

via web interface:

-- drawer -> Files to go into the HDFS user home directory

-- upload the input file

• execute the jar from the gateway: hadoop jar app.jar <number-of-reducers> <input-file> <outputfile>

all the relative path in input and output file are related to the home directory of the user in the hadoop cluster.

The whole job history is available at https://ma1-bigdata.polito.it:19890/jobhistory/, kerberos authentication required: obtain a 1 day ticket for the local machine using kinit s241915

How to submit a Spark job locally or on the cluster

Differently from MapReduce, a spark job can be executed locally or on the hadoop cluster using the YARN scheduler.

To submit a spark application, the spark-submit CLI must be used. It accepts various parameters, values separated using space and not equal

• --class class of the jar to be executed that contains the main
• --master used to specify the scheduler. Values: yarn to execture the application on the cluster, loca to execute on the local workstation.
• --deploy-mode used to specify where the driver must be executed. values: client or cluster. Note that all the local variable of the driver will be istantiated in the driver, in cluster mode the maximum amout of main memry for each executor is 2 to 8GB, better to use client mode if the workstation has more than 8GB of ram.
• --num-executors, default is 2
• --executor-cores, default is 1
• --executor-memory, default is 1GB, main memory for each executor that runs on the cluster
• --driver-cores
• --driver-memory

example:

spark-submit --class it.polito.bigdata.spark.DriverMyApp --deploy-mode cluster --master yarn MyApp.jar <arguments>

example with Eclipse Oxygen and local installation of spark:

1. in Eclipse, File->Export->Jar file, configure and select the main class

2. launch the job locally, from the output directory of the jar:

/opt/spark/bin/spark-submit --class it.polito.bigdata.spark.example.SparkDriver --deploy-mode client --master local lab5_sol.jar <args>

How to Configure Kerberos authentication

A Kerberos ticket is required to have access to other servers of the bigdatalab other that the gateway.

• sudo apt install krb5-user and then follow the guide