Press "Enter" to skip to content

Category: Hadoop

A Modern C++ Kafka API

Published 2021-04-15 by Kevin Feasel

Kenneth Jia and Benedek Thaler announce an open source library:

Morgan Stanley uses Apache Kafka® to publish market data to internal clients and to persist it for replay purposes. We started out using librdkafka’s C++ API, which maintains C++98 compatibility. C++ is evolving quickly, and we wanted to break away from this compatibility requirement so we could take advantage of new C++ features. This led us to create a new C++ API for Kafka that uses modern C++ features (i.e. C++14 and later). We’ve open sourced this client and hope you enjoy it.

Click through to learn more. What interests me about this is that most of the other languages’ support for Kafka (for example, .NET) is based off of librdkafka. I don’t know if there’s any benefit to moving to this new library.

Comments closed

Geospatial Fraud Detection

Published 2021-04-15 by Kevin Feasel

Antoine Amend uses Databricks to identify financial fraud in a geographical area:

As part of this real-world solution, we are releasing a new open source geospatial library, GEOSCAN, to detect geospatial behaviors at massive scale, track customers patterns over time and detect anomalous card transactions. Finally, we demonstrate how organizations can surface anomalies from an analytics environment to an online data store (ODS) with tight SLA requirements following a Lambda-like infrastructure underpinned by Delta Lake, Apache Spark and MLflow.

Click through for the article, as well as three notebooks.

Comments closed

SCD2 Dimensions on Spark with Apache Hudi

Published 2021-04-13 by Kevin Feasel

David Greenshtein shows how we can build type-2 slowly changing dimensions using Apache Hudi:

Implementing SCD2 in a data lake without using an additional framework like Apache Hudi introduces the challenge of updating data stored on immutable Amazon S3 storage, and as a result requires the implementor to create multiple copies of intermediate results. This situation may lead to a significant maintenance effort and potential data loss or data inconsistency.

Apache Hudi is an open-source data management framework used to simplify incremental data processing and data pipeline development. Hudi enables you to manage data at the record level in Amazon S3 and helps to handle data privacy use cases requiring record-level updates and deletes. Hudi is supported by Amazon EMR starting from version 5.28 and is automatically installed when you choose Spark, Hive, or Presto when deploying your EMR cluster.

Click through for the process.

Comments closed

Ordering and Sorting Data in Spark

Published 2021-04-07 by Kevin Feasel

Landon Robinson shows how to sort data in Spark RDDs and DataFrames:

In the analysis section of Spark Starter Guide 4.6: How to Aggregate Data, we asked these questions: “Who is the youngest cat in the data? Who is the oldest?”

Let’s use ordering in Spark as an alternative method to answer those same questions, and achieve the same result. Specifically, let’s again find the youngest and oldest cats in the data.

Click through for plenty of examples.

Comments closed

Sending Large Messages in Kafka

Published 2021-04-06 by Kevin Feasel

The Hadoop in Real World team shows how you can send large messages in Apache Kafka:

By default the messages you can send and manage in Kafka should be less than 1 MB. To increase this limit there are few properties you need to change in both brokers and consumers.

Let’s say your messages can be up to 10 MB. So in this case your Kafka producers are producing messages up to 10 MB. So your Kafka Brokers and consumers should be able to store and receive messages up to 10 MB respectively.

Kafka Producer sends messages up to 10 MB ==> Kafka Broker allows, stores and manages messages up to 10 MB ==> Kafka Consumer receives messages up to 10 MB

Click through to see how, but also recognize that it’s usually a really bad idea to push large messages in a broker system. Even 1MB is probably going too far—I’d try to stay under 1KB if possible.

Comments closed

spkarlyr 1.6 Released

Published 2021-04-06 by Kevin Feasel

Carly Driggers announces a new release of sparklyr:

Sparklyr, an LF AI & Data Foundation Incubation Project, has released version 1.6! Sparklyr is an R Language package that lets you analyze data in Apache Spark, the well-known engine for big data processing, while using familiar tools in R. The R Language is widely used by data scientists and statisticians around the world and is known for its advanced features in statistical computing and graphics. 

Click through to see the changes.

Comments closed

Kafka Sans ZooKeeper

Published 2021-04-05 by Kevin Feasel

Ben Stopford and Ismael Juma give us a preview:

So we’re very pleased to say that the early access of the KIP-500 code has been committed to trunk and is expected to be included in the upcoming 2.8 release. For the first time, you can run Kafka without ZooKeeper. We call this the Kafka Raft Metadata mode, typically shortened to KRaft (pronounced like craft) mode.

Beware, there are some features that are not available in this early-access release. We do not yet support the use of ACLs and other security features or transactions. Also, both partition reassignment and JBOD are unsupported in KRaft mode (these are anticipated to be available in an Apache Kafka release later in the year). Hence, consider the quorum controller experimental software—we don’t advise subjecting it to production workloads. If you do try out the software, however, you’ll find a host of new advantages: It’s simpler to deploy and operate, you can run Kafka in its entirety as a single process, and it can accommodate significantly more partitions per cluster (see measurements below).

Read on for more information. This is a big deal for Kafka.

Comments closed

Caching versus Persisting in Spark

Published 2021-03-31 by Kevin Feasel

The Hadoop in Real World team explains a subtle difference:

cache() and persist() functions are used to cache intermediate results of a RDD or DataFrame or Dataset. You can mark an RDD, DataFrame or Dataset to be persisted using the persist() or cache() methods on it. The first time it is computed in an action, the objects behind the RDD, DataFrame or Dataset on which cache() or persist() is called will be kept in memory or on the configured storage level on the nodes. 

That’s the similarity, but click through for the difference.

Comments closed

Spark Performance in Azure Synapse Analytics

Published 2021-03-31 by Kevin Feasel

Euan Garden shares some numbers around Apache Spark performance in Azure Synapse Analytics:

To compare the performance, we derived queries from TPC-DS with 1TB scale and ran them on 8 nodes Azure E8V3 cluster (15 executors – 28g memory, 4 cores). Even though our version running inside Azure Synapse today is a derivative of Apache Spark™ 2.4.4, we compared it with the latest open-source release of Apache Spark™ 3.0.1 and saw Azure Synapse was 2x faster in total runtime for the Test-DS comparison.

Click through for several techniques the Azure Synapse Analytics team has implemented to make some significant performance improvements. It’s still slower than Databricks, but considerably faster than the open-source Apache Spark baseline.

Comments closed