Apache Kafka wire protocol

Apache Kafka protocol implementations generated for different programming languages (currently, Rust; Go is planned) based on the Kafka code.

The library contains the protocol messages and some convenience code, but this is not a complete client (or server) implementation.

Message versions up to Kafka 3.8.0 are supported at the moment.

Usage

Rust

See rust_examples/ for complete examples.

kafka_wire_protocol::schema namespace contains generated structs for various message types and versions. These structs have read function for reading from Read and write function for writing into Write. This allows reading and writing to/from e.g. Cursor.

let request_header = RequestHeader::new(
    request.api_key(),
    request.version(),
    correlation_id,
    Some(client_id),
);
request_header.write(&mut request_cur).unwrap();

// ...

let response_header = ResponseHeader::read(&mut response_cur).unwrap();

The ApiMessageType struct may be useful for getting various information about versions, e.g. expected header versions.

How it's developed

Apache Kafka's wire protocol is defined as a set of JSON files. There is a Gradle task processMessages which converts these JSON definitions into actual Java classes. This project aims to do the same but for other programming languages. At the moment, its focus is Rust, but Go is also planned. (For Python, have a look at Kio).

This project contains the unaltered code of the clients module and the slightly altered code of generator module from the Apache Kafka code base. The additions are the following:

1. The Rust code (generated and handwritten), located in rust/.
2. The Rust examples, located in rust_examples/.
3. RustMessageGenerator.java and its supporting classes with the Rust prefixes, which contain the actual generation logic.
4. The processMessagesRust Gradle task for running the generator.
5. Java Tester for testing against the real Kafka Java code (see below).

Testing

Serialization/deserialization is a crucial part of the protocol implementation, which must be correct for the library to be useful. The library is tested on several levels:

1. Basic handwritten unit tests.
2. Pure Rust serialization/deserialization property-based tests with proptest.
3. Serialization/deserialization property-based tests that run against the real Java Kafka code. They aim to ensure that the generated Rust code and the upstream Java code understand the protocol completely identical. See the Java Tester section below.
4. Fuzzing tests for deserialization.
5. Integration tests against a real Kafka instance running in Docker.

Testing with Java Tester

Java Tester is a piece of Java code that uses the original Kafka serialization/deserialization classes to check that the generated Rust code produces is correct. Java Tester is taken from Kio.

The Rust test code runs Java Tester and sends test cases in the JSON format to its standard input, expecting the result at the standard output. The result is either a success or a failure with the accompanying error message and/or stack trace.

Test functions use the property-based testing generator to generate random protocol message structures, serialize them in binary and JSON formats, and send them to Java Tester. The latter reconstructs the message based on the JSON representation, serializes it, and compares that the serialized value is identical to what Rust produced, thus ensuring the Rust code does serialization correctly.

See java_tester.rs for details.

Development

Run

make test

to run the generator and execute the tests.

License

This code base--both the forked Kafka code and the original code--is provided under the Apache 2.0 license.