Apache Kafka Three Ways
An Open Source Apache Kafka Workshop provided by Troy-West.
By developers, for developers, our workshop takes new starters through the nuts and bolts of Kafka. We explain the big ideas, layered abstractions, and gradual evolution of Kafka over the years.
Morning materials provided as a reveal.js presentation and Q+A (theory).
We offer a guided, full day workshop with your team - or use and adapt these MIT licensed materials as you see fit.
To learn the fundamentals, ergonomics, trade-offs, and maturity of Apache Kafka as:
- A Message Broker.
- A Streaming Compute Platform.
- A Distributed Database.
To build a streaming compute application that processes +1M messages and solves the mystery of the Number Stations.
To operate the streaming compute application with a local Kafka cluster, understanding the tiered abstractions of Apache Kafka and how they enable availability, scalability, and (near) real-time compute.
Time permitting, we dabble with vendor tooling (KSQL, Kafka-Connect).
- Java 11+ or Clojure 10+ and an IDE for completing the Number Stations project work
- Docker & Docker Compose installed
- troy-west/apache-kafka-cli-tools cloned
- troy-west/apache-kafka-number-stations (Java) cloned
- or, troy-west/apache-kafka-number-stations-clj (Clojure) cloned
- troy-west/apache-kafka-vendor-tools cloned
Completed solutions to both the Java and Clojure programming exercises are available:
- troy-west/apache-kafka-number-stations-sln (Java)
- troy-west/apache-kafka-number-stations-clj-sln (Clojure)
Morning Session (Theory)
A deep dive into the history and fundamentals of Apache Kafka as a Message Broker.
We learn about Kafka via the lens of three key project decisions and their trade-offs:
- High Availability
- Linear Scalability
- Near Real-Time Compute
Throughout we focus on real-time data's unifying abstraction, the log.
After the presentation and Q+A we use troy-west/apache-kafka-cli-tools to start a local, 3-node Kafka Cluster. We operate that cluster via the shell scripts provided by Kafka, exploring repartitioning, offsets, consumer groups, and more.
Afternoon Session (Practice)
Solve the mystery of the Numbers Stations!
We talk briefly about the big ideas behind the Processor API and Kafka Streams, and how Kafka facilitates highly available, linearly scalable, near real-time compute. We share our experience with Apache Storm in this space, and comment on similarities to Clojure regarding immutable streams of data and functional composition.
Then we introduce the mystery of troy-west/apache-kafka-number-stations (or the Clojure variant).
We inspect the secret radio, send 1.5M messages to local Kafka, then progressively fix unit-tests until we have built a streaming compute application that filters, branches, maps, groups, windows, and aggregates data - decoding the secret message and solving the mystery!
Once the tests are green we build and deploy multiple versions of the system, discussing the expected and observed impact on local state partitioning.
Finally we discuss Interactive Queries and the possibility of using Kafka as a distributed database. What data should we expect to source from Kakfa, just the logs, or derived / materialized / computed data?
If you are experienced with Kafka you might find these workshop materials useful for your own team.
Start with the index.html in this project, or view at kafka.troywest.com.
The presentation moves left to right, then vertically for each the broker, compute, db, and vendor parts.
Copyright © 2019 Troy-West, Pty Ltd. MIT Licensed. Contributions welcome.