Kafka: The Definitive Guide went to print in December 2021, more than three years before Kafka 4.0 removed ZooKeeper. It still had the best Amazon sales rank of the Kafka books here when we checked, which means many readers start on a pre-4.0 book. Spark has the same gap, because Learning Spark covers Spark 3.0 while Spark 4.0 turned on ANSI SQL mode and made JDK 17 the baseline.
These twelve Apache Kafka and Spark books take you from your first topic to a tuned Spark job, with Flink and Beam picks for stream processing. You’ll see which release each book targets where the publisher says, and what changed since. Editions, prices and stock come from Amazon’s live catalog and contents from each publisher’s table of contents, all checked in October 2026.
Pick a Kafka or Spark book by your role
Most readers need one Kafka book and one Spark book, not twelve. Start from the job you do with them:
- Running Kafka clusters: Apache Kafka in Action (Zelenin, Kropp)
- Kafka internals and reference: Kafka: The Definitive Guide, 2nd Edition (Shapira and others)
- Writing stream processing apps on Kafka: Kafka Streams in Action, 2nd Edition (Bill Bejeck)
- Designing systems around Kafka: Kafka for Architects (Katya Gorshkova)
- Tuning Spark 4 jobs: High Performance Spark, 2nd Edition (Karau, Polak, Warren)
- Learning Spark from Python: Data Analysis with Python and PySpark (Jonathan Rioux)
If you want the theory under all of it, add Streaming Systems for event time and watermarks, or the second edition of Designing Data-Intensive Applications for the storage and replication ideas Kafka is built on. Before you open a Kafka book, it helps to have a broker to poke at: our Kafka Docker Compose guide runs the broker in KRaft mode, first as one node and then as three.
How these books were chosen
A plain Amazon search for “Kafka” returns Franz Kafka’s novels before it returns a single data book, and the Flink and Beam shelves are mostly short self-published titles. We worked from the publishers’ catalogs instead, querying O’Reilly, Manning, Packt and Apress for Kafka, Spark, Flink and stream processing, and kept a book only if it passed three checks:
- an established technical publisher
- a print edition that was buyable new when we checked
- a table of contents or publisher description we could read
We didn’t read all twelve cover to cover, so this is a buyer’s guide, not twelve full reviews. Each book was judged on its publisher description, its table of contents, its companion code where a public repository exists, and how far the release it teaches sits from today’s. We opened the companion repositories for Kafka Streams in Action and Stream Processing with Apache Flink, and what they pin is noted in those two sections. Prices are what Amazon charged on the day we checked, so treat them as a guide.
Compare the Kafka and Spark books
Read the publication date against the two big releases first. Kafka 4.0 shipped in March 2025 and Spark 4.0 in May 2025, so anything older was written for the previous major line:
| Book | Author / Publisher | Published | Pages | Focus | Price when checked |
|---|---|---|---|---|---|
| Kafka: The Definitive Guide, 2nd ed. | Shapira et al. / O’Reilly | Dec 2021 | 485 | Kafka internals, operations | $44 |
| Apache Kafka in Action | Zelenin, Kropp / Manning | Jun 2025 | 368 | Running Kafka in production | $42 |
| Kafka Streams in Action, 2nd ed. | Bill Bejeck / Manning | May 2024 | 504 | Kafka Streams, Connect, ksqlDB | $51 |
| Kafka for Architects | Katya Gorshkova / Manning | Feb 2026 | 392 | Architecture, data contracts | $58 |
| High Performance Spark, 2nd ed. | Karau, Polak, Warren / O’Reilly | Jul 2026 | 409 | Spark 4.x tuning | $50 |
| Learning Spark, 2nd ed. | Damji et al. / O’Reilly | Aug 2020 | 397 | Spark 3.0 Structured APIs | $44 |
| Data Analysis with Python and PySpark | Jonathan Rioux / Manning | Mar 2022 | 456 | PySpark for Python users | $60 |
| Spark: The Definitive Guide | Chambers, Zaharia / O’Reilly | Apr 2018 | 603 | Spark 2.0 reference | $42 |
| Streaming Systems | Akidau, Chernyak, Lax / O’Reilly | Aug 2018 | 349 | Stream processing theory | $63 |
| Stream Processing with Apache Flink | Hueske, Kalavri / O’Reilly | Apr 2019 | 308 | Flink DataStream API | $42 |
| Building Big Data Pipelines with Apache Beam | Jan Lukavsky / Packt | Jan 2022 | 342 | Beam in Java | $46 |
| Designing Data-Intensive Applications, 2nd ed. | Kleppmann, Riccomini / O’Reilly | Mar 2026 | 670 | Data systems design | $57 |
Four of the twelve came out after the Kafka 4.0 and Spark 4.0 releases. Being older does not make a book wrong: the log and partition model at the heart of Kafka is the same in 4.0, and Spark’s DataFrame API reads much the same in 3.0 and 4.x. What changed is the setup and the defaults, which the section at the end lists so you can correct for it while you read.
1. Kafka: The Definitive Guide
Buy this as your Kafka reference. Engineers from Confluent and LinkedIn wrote it, and the second edition is the one to get: no third edition was out or announced when we checked.
It explains how Kafka works underneath, which is what you need once a cluster starts misbehaving. O’Reilly lists the replication protocol, the controller and the storage layer among the internals it covers, and the second edition added chapters on the AdminClient API, transactions, newer security features and changed tooling. The operations half covers deployment, configuration, monitoring, tuning and the metrics worth alerting on, which pairs well with our walkthrough to monitor Kafka with Prometheus and Grafana.
Its weak spot is age. It was published in December 2021, before Kafka 4.0 removed ZooKeeper and made KRaft the only mode, so check any cluster setup steps against the current Kafka documentation rather than copying them. Amazon had it at about $44 (list $79.99) when we checked. Check the current price on Amazon.
2. Apache Kafka in Action
Pick this one if you are the person who has to keep Kafka running. Anatoly Zelenin and Alexander Kropp wrote it for operators, architects and developers with no Kafka experience, and Amazon lists the print edition from June 2025, after Kafka 4.0 shipped.
The eighteen chapters move from first steps to the work that fills an operator’s week. The second half is where it earns its place on the list:
- cluster management, producing, consuming and cleaning up messages
- Kafka Connect and stream processing
- governance and a reference architecture
- monitoring and alerting, plus disaster management
- a chapter comparing Kafka with other technologies
Two appendices set up a test environment and a monitoring stack, so you can follow along on a lab box. If you’d rather start on a plain server, our guide to install Kafka on Ubuntu gets you a single broker to practice on.
Stream processing gets one chapter, so developers who plan to write stream processing code will want the next pick as well. When we checked, the Amazon listing showed a third-party seller and a few days’ lead time. It was about $42 (list $59.99). Check the current price on Amazon.
3. Kafka Streams in Action
Developers should start here. Bill Bejeck, a Confluent engineer and Apache Kafka committer, rewrote the second edition to cover the whole platform around Kafka Streams, not just the library.
Its fourteen chapters start with the platform around the library, which is why you can buy it without knowing Kafka first. The first five chapters cover brokers, Schema Registry, the producer and consumer clients and Kafka Connect, and the rest of the book is stream processing:
- state stores, the KTable API, windowing and timestamps
- the Processor API for low-level control
- ksqlDB and Spring Kafka
- interactive queries and testing
It’s written for Java developers and assumes no Kafka knowledge, so you can open it before you have ever run a broker.
The companion repository was still being updated in May 2026, and its Docker Compose file runs the broker in KRaft mode with the broker,controller process roles, so the examples do not depend on ZooKeeper even though the book shipped in May 2024.
Skip it if you don’t write Java, because the examples are JVM code throughout. The Amazon listing showed a third-party seller with limited stock when we checked. It was about $51 (list $59.99). Check the current price on Amazon.
4. Kafka for Architects
Get this if you decide where Kafka fits rather than how it is configured. Katya Gorshkova, a consulting engineer and architect in the FinTech sector, kept it deliberately above the code.
The twelve chapters cover cluster data architecture, clients, data contracts and schema evolution, interaction patterns such as CQRS and event sourcing, and the organisational side: running a Kafka project and operating Kafka inside an enterprise. Published in February 2026, it is the newest Kafka book on this list.
For a free companion on the same theme, Confluent gives away Designing Event-Driven Systems as an ebook, which covers event sourcing and replayable logs with Kafka.
If you want to build something this week, look elsewhere: there’s little code, by design. Amazon sold it at about $58 (list $69.99) when we checked. Check the current price on Amazon.
5. High Performance Spark
If you already run Spark, buy this one. The second edition came out in July 2026, and O’Reilly describes it as written for Spark 4.x and beyond, so it is the one pick here that matches the Spark you’d install today.
Holden Karau, Adi Polak and Rachel Warren focus on why jobs run slowly and how to fix them. O’Reilly lists key skew, Spark’s dynamic partitioning, scalable testing and validation, Spark on Kubernetes, GPU acceleration with RAPIDS and resource profiles, and PyTorch integration among the topics.
If you need a cluster to tune, our steps to install Apache Spark on Debian or Ubuntu give you one to practice against.
Beginners should hold off and start with Learning Spark or the PySpark book, because this one assumes you’ve already written Spark jobs and want them faster. It was about $50 (list $65.99) when we checked. Check the current price on Amazon.
6. Learning Spark
Learning Spark teaches Spark’s Structured APIs from the ground up. Four people from Databricks wrote it, and Matei Zaharia, who created Spark, wrote the foreword.
It teaches DataFrames and Spark SQL in Python, SQL, Scala and Java, then moves on to the Spark UI, Structured Streaming, Delta Lake, MLlib and MLflow. The second edition was updated for Spark 3.0, and the core of what it teaches, the DataFrame and SQL model, still carries over to Spark 4.
A third edition updated for Spark 4.0 is on preorder, with Amazon showing a June 2027 date when we checked. If you’re starting now, buy the second edition and read the Spark 4.0 release notes alongside it.
It predates Spark 4.0. Spark 3.0 ran with ANSI SQL mode off by default, so an example that relies on a NULL from an invalid cast will raise an error on Spark 4 instead. Stock was limited when we checked. It was about $44 (list $79.99). Check the current price on Amazon.
7. Data Analysis with Python and PySpark
Choose this over Learning Spark if you live in Python and pandas. Jonathan Rioux, who uses PySpark daily as an ML director, wrote it for data scientists and data engineers, and it assumes no Spark knowledge.
The fourteen chapters build from a first PySpark program to the parts that go beyond pandas: blending Python and SQL, RDDs and UDFs, pandas UDFs, window functions and reading Spark’s query plans. The last three chapters cover machine learning pipelines, including custom transformers and estimators.
Its examples are PySpark only, so Scala and Java teams should pick Learning Spark instead. Manning does not state the Spark version it targets, and it was published in March 2022, before Spark 4. Amazon had it at about $60 when we checked. Check the current price on Amazon.
8. Spark: The Definitive Guide
Bill Chambers and Matei Zaharia wrote the big Spark reference, and at 603 pages it’s the longest Spark book on this list, covering both the structured and low-level APIs.
O’Reilly’s description puts the emphasis on Spark 2.0. It covers DataFrames, SQL and Datasets, RDDs and how SQL is executed, running on a cluster, monitoring and tuning, Structured Streaming and MLlib.
Leave it on the shelf unless you want depth over currency: it was published in April 2018 and predates both Spark 3 and Spark 4, so expect older defaults and Spark 2 era build setup. Stock was limited when we checked. It was about $42 (list $69.99). Check the current price on Amazon.
9. Streaming Systems
Read this when you want to understand stream processing rather than one tool. Tyler Akidau, Slava Chernyak and Reuven Lax expanded it from Akidau’s “Streaming 101” and “Streaming 102” posts, and it is written to be platform agnostic.
It covers out-of-order data, watermarks, exactly-once processing, how streams and tables relate, and how time-varying relations connect streaming to SQL. Those ideas sit under Kafka Streams, Flink and Beam alike, so it ages more slowly than a tool book.
It will not teach you an API. If you need working code this month, buy a tool book first and come back to this one. Stock was limited when we checked. It was about $63 (list $79.99). Check the current price on Amazon.
10. Stream Processing with Apache Flink
Fabian Hueske and Vasiliki Kalavri, two longtime Flink committers, wrote the only full-length Flink book from a major publisher that was still buyable new when we checked. Packt’s 2017 Learning Apache Flink was out of stock, and O’Reilly’s 2016 Introduction to Apache Flink is a 107-page primer.
It explains Flink’s architecture, event-time processing and fault-tolerance model, then teaches the DataStream API, exactly-once connectors and how to run a Flink cluster. The concepts carry over to current Flink.
The code does not. The book’s Java example repository pins Flink 1.7.1, and its sensor source extends RichParallelSourceFunction. Flink 2.0, released in March 2025, removed the old SourceFunction interface and the Scala DataStream API, so those examples need porting to the new Source API before they’ll build on Flink 2.
Buy it for the model, not the snippets. It was published in April 2019 and stock was limited when we checked. It was about $42 (list $79.99). Check the current price on Amazon.
11. Building Big Data Pipelines with Apache Beam
Jan Lukavsky’s Packt book is the Beam pick, aimed at engineers who want one programming model for batch and streaming jobs.
Its eight chapters cover basic pipelines and how to test them, stateful processing with state and timers, structuring reusable code, Beam SQL, the Python SDK through Beam’s portability layer, custom I/O connectors with the Splittable DoFn API, and how runners execute a pipeline.
It assumes intermediate Java, so Python-first teams get only one chapter in their own language. Packt does not state a Beam version, and it came out in January 2022. Amazon had it at about $46 (list $50.99) when we checked. Check the current price on Amazon.
12. Designing Data-Intensive Applications
This is not a Kafka or Spark book, but it explains the replication, partitioning and log ideas both are built on.
Martin Kleppmann and Chris Riccomini’s second edition came out in early 2026, the ebook in February and print in March. Its chapters on replication and sharding (the second edition’s name for partitioning) are the background that makes Kafka’s guarantees make sense. It also sits on our list of the best system design books.
Skip it if you only need to operate one tool, because it teaches concepts, not commands. It was about $57 (list $69.99) when we checked. Check the current price on Amazon.
What the books miss about Kafka 4 and Spark 4
Every Kafka book here except Kafka for Architects and Apache Kafka in Action was published before Kafka 4.0, and every Spark book except High Performance Spark predates Spark 4.0. These are the changes to correct for while you read, taken from the projects’ own release notes:
| Release | What changed | What to do while reading an older book |
|---|---|---|
| Kafka 4.0 (March 2025) | ZooKeeper removed; KRaft is the only mode | Skip ZooKeeper setup steps and configure KRaft controllers instead |
| Kafka 4.0 | New consumer group protocol (KIP-848) generally available | Older books describe only the classic rebalance protocol |
| Kafka 4.0 | Brokers, Connect and tools need Java 17; clients and Streams need Java 11 | Install JDK 17 on broker hosts |
| Spark 4.0 (May 2025) | ANSI SQL mode on by default | Expect errors where Spark 3 returned NULL, for example on invalid casts |
| Spark 4.0 | Scala 2.13 and JDK 17 are the defaults; Scala 2.12 and JDK 8/11 dropped | Rebuild Scala examples against 2.13 and run on JDK 17 |
| Flink 2.0 (March 2025) | SourceFunction, SinkFunction and the Scala DataStream API removed | Port book examples to the Source and Sink V2 APIs in Java |
None of this makes the older books obsolete. The way Kafka stores and replicates a partition, and the way Spark plans and runs a query, are untouched by the changes in that table. Read the concepts from whichever book suits you, then follow the current documentation for installation and defaults. If you’re still choosing a broker, our Kafka vs Redpanda benchmarks compare the two on identical VMs, and the cloud computing book guide and programming book list cover the ground around these tools.
