Overview

Introduction to Spark on HDInsight

This article provides you with an introduction to Spark on HDInsight. Apache Spark is an open-source parallel processing framework that supports in-memory processing to boost the performance of big-data analytic applications. Spark cluster on HDInsight is compatible with Azure Storage (WASB) as well as Azure Data Lake Store. Hence, your existing data stored in Azure can easily be processed via a Spark cluster.

When you create a Spark cluster on HDInsight, you create Azure compute resources with Spark installed and configured. It only takes about 10 minutes to create a Spark cluster in HDInsight. The data to be processed is stored in Azure Storage or Azure Data Lake Store. See Use Azure Storage with HDInsight.

Spark vs. traditional MapReduce

What makes Spark fast? How is the architecture of Apache Spark different than traditional MapReduce, allowing it to offer better performance for data sharing?

Spark provides primitives for in-memory cluster computing. A Spark job can load and cache data into memory and query it repeatedly, much more quickly than disk-based systems. Spark also integrates into the Scala programming language to let you manipulate distributed data sets like local collections. There’s no need to structure everything as map and reduce operations.

In Spark, data sharing between operations is faster since data is in-memory. In contrast, Hadoop shares data through HDFS, which takes longer to process.

What is Apache Spark on Azure HDInsight?

Spark clusters on HDInsight offer a fully managed Spark service. Benefits of creating a Spark cluster on HDInsight are listed here.

Spark cluster architecture

Here is the Spark cluster architecture and how it works:

The head node has the Spark master that manages the number of applications, the apps are mapped to the Spark driver. Every app is managed by Spark master in various ways. Spark can be deployed on top of Mesos, YARN, or the Spark cluster manager, which allocates worker node resources to an application. In HDInsight, Spark runs using the YARN cluster manager. The resources in the cluster are managed by Spark master in HDInsight. That means the Spark master has knowledge of which resources, like memory, are occupied or available on the worker node.

The driver runs the user’s main function and executes the various parallel operations on the worker nodes. Then, the driver collects the results of the operations. The worker nodes read and write data from and to the Hadoop distributed file system (HDFS). The worker nodes also cache transformed data in-memory as Resilient Distributed Datasets (RDDs).

Once the app is created in the Spark master, the resources are allocated to the apps by Spark master, creating an execution called the Spark driver. The Spark driver also creates the SparkContext and also starts creating the RDDs. The metadata of the RDDs are stored on the Spark driver.

The Spark driver connects to the Spark master and is responsible for converting an application to a directed graph (DAG) of individual tasks that get executed within an executor process on the worker nodes. Each application gets its own executor processes, which stay up for the duration of the whole application and run tasks in multiple threads.

What are the use cases for Spark on HDInsight?

Spark clusters in HDInsight enable the following key scenarios:

Interactive data analysis and BI

Look at a tutorial

Apache Spark in HDInsight stores data in Azure Storage or Azure Data Lake Store. Business experts and key decision makers can analyze and build reports over that data and use Microsoft Power BI to build interactive reports from the analyzed data. Analysts can start from unstructured/semi structured data in cluster storage, define a schema for the data using notebooks, and then build data models using Microsoft Power BI. Spark clusters in HDInsight also support a number of third-party BI tools such as Tableau making it an ideal platform for data analysts, business experts, and key decision makers.

Spark Machine Learning

Look at a tutorial: Predict building temperatures using HVAC data

Look at a tutorial: Predict food inspection results

Apache Spark comes with MLlib, a machine learning library built on top of Spark that you can use from a Spark cluster in HDInsight. Spark cluster on HDInsight also includes Anaconda, a Python distribution with a variety of packages for machine learning. Couple this with a built-in support for Jupyter and Zeppelin notebooks, and you have a top-of-the-line environment for creating machine learning applications.

Spark streaming and real-time data analysis

Look at a tutorial

Spark clusters in HDInsight offer a rich support for building real-time analytics solutions. While Spark already has connectors to ingest data from many sources like Kafka, Flume, Twitter, ZeroMQ, or TCP sockets, Spark in HDInsight adds first-class support for ingesting data from Azure Event Hubs. Event Hubs is the most widely used queuing service on Azure. Having an out-of-the-box support for Event Hubs makes Spark clusters in HDInsight an ideal platform for building real time analytics pipeline.

What components are included as part of a Spark cluster?

Spark clusters in HDInsight include the following components that are available on the clusters by default.

• Spark Core. Includes Spark Core, Spark SQL, Spark streaming APIs, GraphX, and MLlib.
• Anaconda
• Livy
• Jupyter notebook
• Zeppelin notebook

Spark clusters on HDInsight also provide an ODBC driver for connectivity to Spark clusters in HDInsight from BI tools such as Microsoft Power BI and Tableau.

Source: Introduction to Spark on Azure HDInsight | Microsoft Docs