Apache Spark on HAQM EMR

EMR features HAQM EMR runtime for Apache Spark, a performance-optimized runtime environment for Apache Spark that is active by default on HAQM EMR clusters. HAQM EMR runtime for Apache Spark can be over 3x faster than clusters without the EMR runtime, and has 100% API compatibility with standard Apache Spark. This improved performance means your workloads run faster and saves you compute costs, without making any changes to your applications.

By using a directed acyclic graph (DAG) execution engine, Spark can create efficient query plans for data transformations. Spark also stores input, output, and intermediate data in-memory as resilient dataframes, which allows for fast processing without I/O cost, boosting performance of iterative or interactive workloads.

Apache Spark natively supports Java, Scala, SQL, and Python, which gives you a variety of languages for building your applications. Also, you can submit SQL or HiveQL queries using the Spark SQL module. In addition to running applications, you can use the Spark API interactively with Python or Scala directly in the Spark shell or via EMR Studio, or Jupyter notebooks on your cluster. Support for Apache Hadoop 3.0 in EMR 6.0 brings Docker container support to simplify managing dependencies. You can also leverage cluster-independent EMR Notebooks (based on Jupyter) or use Zeppelin to create interactive and collaborative notebooks for data exploration and visualization. You can tune and debug your workloads in the EMR console which has an off-cluster, persistent Spark History Server.

Apache Spark includes several libraries to help build applications for machine learning (MLlib), stream processing (Spark Streaming), and graph processing (GraphX). These libraries are tightly integrated in the Spark ecosystem, and they can be leveraged out of the box to address a variety of use cases. Additionally, you can use deep learning frameworks like Apache MXNet with your Spark applications. Integration with AWS Step Functions enables you to add serverless workflow automation and orchestration to your applications.

Submit Apache Spark jobs with the EMR Step API, use Spark with EMRFS to directly access data in S3, save costs using EC2 Spot capacity, use EMR Managed Scaling to dynamically add and remove capacity, and launch long-running or transient clusters to match your workload. You can also easily configure Spark encryption and authentication with Kerberos using an EMR security configuration. Additionally, you can use the AWS Glue Data Catalog to store Spark SQL table metadata or use HAQM SageMaker with your Spark machine learning pipelines. EMR installs and manages Spark on Hadoop YARN, and you can also add other big data applications on your cluster. EMR with Apache Hudi lets you more efficiently manage change data capture (CDC) and helps with privacy regulations like GDPR and CCPA by simplifying record deletion. Click here for more details about EMR features.

Consume and process real-time data from HAQM Kinesis, Apache Kafka, or other data streams with Spark Streaming on EMR. Perform streaming analytics in a fault-tolerant way and write results to S3 or on-cluster HDFS.

Apache Spark on EMR includes MLlib for a variety of scalable machine learning algorithms, or you can use your own libraries. By storing datasets in-memory during a job, Spark has great performance for iterative queries common in machine learning workloads. You can enhance HAQM SageMaker capabilities by connecting the notebook instance to an Apache Spark cluster running on HAQM EMR, with HAQM SageMaker Spark for easily training models and hosting models.

Use Spark SQL for low-latency, interactive queries with SQL or HiveQL. Spark on EMR can leverage EMRFS, so you can have ad hoc access to your datasets in S3. Also, you can utilize EMR Studio, EMR Notebooks, Zeppelin notebooks, or BI tools via ODBC and JDBC connections.