Home

Awesome

rust-rdkafka

crates.io docs.rs Build Status coverate Join the chat at https://gitter.im/rust-rdkafka/Lobby

A fully asynchronous, futures-enabled Apache Kafka client library for Rust based on librdkafka.

The library

rust-rdkafka provides a safe Rust interface to librdkafka. This version is compatible with librdkafka v1.9.2+.

Documentation

Features

The main features provided at the moment are:

One million messages per second

rust-rdkafka is designed to be easy and safe to use thanks to the abstraction layer written in Rust, while at the same time being extremely fast thanks to the librdkafka C library.

Here are some benchmark results using the BaseProducer, sending data to a single Kafka 0.11 process running in localhost (default configuration, 3 partitions). Hardware: Dell laptop, with Intel Core i7-4712HQ @ 2.30GHz.

For more numbers, check out the kafka-benchmark project.

Client types

rust-rdkafka provides low level and high level consumers and producers.

Low level:

High level:

For more information about consumers and producers, refer to their module-level documentation.

Warning: the library is under active development and the APIs are likely to change.

Asynchronous data processing with Tokio

Tokio is a platform for fast processing of asynchronous events in Rust. The interfaces exposed by the StreamConsumer and the FutureProducer allow rust-rdkafka users to easily integrate Kafka consumers and producers within the Tokio platform, and write asynchronous message processing code. Note that rust-rdkafka can be used without Tokio.

To see rust-rdkafka in action with Tokio, check out the asynchronous processing example in the examples folder.

At-least-once delivery

At-least-once delivery semantics are common in many streaming applications: every message is guaranteed to be processed at least once; in case of temporary failure, the message can be re-processed and/or re-delivered, but no message will be lost.

In order to implement at-least-once delivery the stream processing application has to carefully commit the offset only once the message has been processed. Committing the offset too early, instead, might cause message loss, since upon recovery the consumer will start from the next message, skipping the one where the failure occurred.

To see how to implement at-least-once delivery with rdkafka, check out the at-least-once delivery example in the examples folder. To know more about delivery semantics, check the message delivery semantics chapter in the Kafka documentation.

Exactly-once semantics

Exactly-once semantics (EOS) can be achieved using transactional producers, which allow produced records and consumer offsets to be committed or aborted atomically. Consumers that set their isolation.level to read_committed will only observe committed messages.

EOS is useful in read-process-write scenarios that require messages to be processed exactly once.

To learn more about using transactions in rust-rdkafka, see the Transactions section of the producer documentation.

Users

Here are some of the projects using rust-rdkafka:

If you are using rust-rdkafka, please let us know!

Installation

Add this to your Cargo.toml:

[dependencies]
rdkafka = { version = "0.25", features = ["cmake-build"] }

This crate will compile librdkafka from sources and link it statically to your executable. To compile librdkafka you'll need:

Note that using the CMake build system, via the cmake-build feature, is encouraged if you can take the dependency on CMake.

By default a submodule with the librdkafka sources pinned to a specific commit will be used to compile and statically link the library. The dynamic-linking feature can be used to instead dynamically link rdkafka to the system's version of librdkafka. Example:

[dependencies]
rdkafka = { version = "0.25", features = ["dynamic-linking"] }

For a full listing of features, consult the rdkafka-sys crate's documentation. All of rdkafka-sys features are re-exported as rdkafka features.

Minimum supported Rust version (MSRV)

The current minimum supported Rust version (MSRV) is 1.70.0. Note that bumping the MSRV is not considered a breaking change. Any release of rust-rdkafka may bump the MSRV.

Asynchronous runtimes

Some features of the StreamConsumer and FutureProducer depend on Tokio, which can be a heavyweight dependency for users who only intend to use the low-level consumers and producers. The Tokio integration is enabled by default, but can be disabled by turning off default features:

[dependencies]
rdkafka = { version = "0.25", default-features = false }

If you would like to use an asynchronous runtime besides Tokio, you can integrate it with rust-rdkafka by providing a shim that implements the AsyncRuntime trait. See the following examples for details:

Examples

You can find examples in the examples folder. To run them:

cargo run --example <example_name> -- <example_args>

Debugging

rust-rdkafka uses the log crate to handle logging. Optionally, enable the tracing feature to emit tracing events as opposed to log records.

In test and examples, rust-rdkafka uses the env_logger crate to format logs. In those contexts, logging can be enabled using the RUST_LOG environment variable, for example:

RUST_LOG="librdkafka=trace,rdkafka::client=debug" cargo test

This will configure the logging level of librdkafka to trace, and the level of the client module of the Rust client to debug. To actually receive logs from librdkafka, you also have to set the debug option in the producer or consumer configuration (see librdkafka configuration).

To enable debugging in your project, make sure you initialize the logger with env_logger::init(), or the equivalent for any log-compatible logging framework.

rdkafka-sys

See rdkafka-sys.

Contributors

Thanks to:

Alternatives