Spark

The question about why Apache Spark is "slow" is one of the most often questions I'm hearing from junior engineers and peoples I'm mentoring. While that is partially true, it should be clarified. TLDR – OSS Spark is a multi-purpose engine that is designed to handle different kinds of workloads. Under the hood of Spark is using a data-centric code generation but also it has some vectorization as well as option to fallbak to a pure Volcano-mode. Because of that Spark can be considred as a hybrid engine, that can benefit from all the approaches. But because of it's multi-purpose nature it will be almost always slower compared to pure vectorized engines like Trino on OLAP workloads on top of columnar data, except rare cases of big amount of nulls or deep branching in the query. In this blogpost I'm trying to explain the statement above.

In this blog post, I will provide a brief high-level overview of projects designed to accelerate Apache Spark by the native physical execution, including Databricks Photon, Apache Datafusion Comet, and Apache Gluten (incubating). I will explain the problems these projects aim to solve and their approaches. The main focus will be on the Comet project, particularly its internal architecture. Additionally, I will share my personal experience of making my first significant contribution to the project. This will include not only a description of the problem I solved and my solution but also insights into the overall contribution experience and the pull request review process.

Summary This blog post is a detailed story about how I ported a popular data quality framework, AWS Deequ, to Spark-Connect. Deequ is a very cool, reliable and scalable framework that allows to compute a lot of metrics, checks and anomaly detection suites on the data using Apache Spark cluster. But the Deequ core is a Scala library that uses a lot of low-level Apache Spark APIs for better performance, so it cannot be run directly on any of Spark-Connect environment. To solve this problem, I defined protobuf messages for all main structures of Deequ, like Check, Analyzer, AnomalyDetectionStrategy, etc., wrote a helper object that can re-create Deequ structures from the corresponding protobuf, and finally made a Spark-Connect native plugin that can process Deequ specific messages, construct DQ suits from them, compute the report, and return the result to the Spark-Connect client. I tested my solution with PySpark Connect 3.5.1, but it should work with any of the existing Spark-Connect clients (Spark-Connect Java/Scala, Spark-Connect Go, Spark-Connect Rust, Spark-Connect C#, etc). ...

This blog post presents a very detailed step-by-step guide on how to create a SparkConnect protocol extension in Java and call it from PySpark. It will also cover a topic about how to define all the necessary proto3 messages for it. At the end of this guide you will have a way to interact with Spark JVM from PySpark almost like you can with py4j in a non-connect version.

I recently had the opportunity to work on an open source project that implements a custom Apache Spark data source and associated logic for working with graph data. The code was written to work with Apache Spark 3.2.2. I am committed to extending support to multiple versions of Spark. In this blog post I want to show how the structure of such a project can be organized using Maven profiles.

In this post, I want to discuss the groundbreaking changes in the latest LTS release of the Databricks runtime. This release introduced Spark Connect as the default way to work with shared clusters. I will give a brief introduction to the topic of internal JVM calls and Spark Connect, provide examples of 3d-party OSS projects broken in 14.3, and try to understand the reasons for such a move by Databricks.

In this post, I will show you how to use information from the spark plan to track data lineage at the column level. This approach will also works with recently introduced SparkConnect.

Sometimes it is an important question, how much memory does our DataFrame use? And there is no easy answer if you are working with PySpark. You can try to collect the data sample and run local memory profiler. You can estimate the size of the data in the source (for example, in parquet file). But we will go another way and try to analyze the logical plan of Spark from PySpark. In case when we are working with Scala Spark API we are able to work with resolved or unresolved logical plans and physical plan via a special API. But from PySpark API only string representation is available and we will work with it.

Working with File System from PySpark Motivation Any of us is working with File System in our work. Almost every pipeline or application has some kind of file-based configuration. Typically json or yaml files are used. Also for data pipelines, it is sometimes important to be able to write results or state them in a human-readable format. Or serialize some artifacts, like matplotlib plot, into bytes and write them to the disk. ...