Baike.dev
All toolsOnline toolsAI codingTrendingOpen sourceNewsSubmit
Log in
< Back to tools
R

rskafka

> 编程语言
Open source

A minimal Rust client for Apache Kafka

336 stars0 likes0 views
WebsiteGitHub

About

A minimal Rust client for Apache Kafka

RSKafka

This crate aims to be a minimal Kafka implementation for simple workloads that wish to use Kafka as a distributed write-ahead log.

It is not a general-purpose Kafka implementation, instead it is heavily optimised for simplicity, both in terms of implementation and its emergent operational characteristics. It was originally used by InfluxDB 3.0 but no longer is.

This crate has:

  • No support for offset tracking, consumer groups, transactions, etc...
  • No built-in buffering, aggregation, linger timeouts, etc...
  • Independent write streams per partition

It will be a good fit for workloads that:

  • Perform offset tracking independently of Kafka
  • Read/Write reasonably sized payloads per-partition
  • Have a low number of high-throughput partitions [^1]

Usage

…

For more advanced production and consumption, see [crate::client::producer] and [crate::client::consumer].

Features

  • compression-gzip (default): Support compression and decompression of messages using gzip.
  • compression-lz4 (default): Support compression and decompression of messages using LZ4.
  • compression-snappy (default): Support compression and decompression of messages using Snappy.
  • compression-zstd (default): Support compression and decompression of messages using zstd.
  • full: Includes all stable features (compression-gzip, compression-lz4, compression-snappy, compression-zstd, transport-socks5, transport-tls).
  • transport-socks5: Allow transport via SOCKS5 proxy.
  • transport-tls: Allows TLS transport via rustls using the ring crypto provider.
  • transport-tls-aws-lc-rs: Allows TLS transport via rustls using the aws-lc-rs crypto provider.
  • transport-tls-no-provider: Allows TLS transport via rustls without enabling a built-in crypto provider.
  • transport-tls-ring: Allows TLS transport via rustls using the ring crypto provider.
  • unstable-fuzzing: Exposes some internal data structures so that they can be used by our fuzzers. This is NOT a stable feature / API!

Testing

Redpanda

To run integration tests against Redpanda, run:

bash
$ docker-compose -f docker-compose-redpanda.yml up

in one session, and then run:

bash
$ TEST_INTEGRATION=1 TEST_BROKER_IMPL=redpanda KAFKA_CONNECT=0.0.0.0:9011 cargo test

in another session.

Apache Kafka

To run integration tests against Apache Kafka, run:

bash
$ docker-compose -f docker-compose-kafka.yml up

in one session, and then run:

bash
$ TEST_INTEGRATION=1 TEST_BROKER_IMPL=kafka KAFKA_CONNECT=localhost:9011 KAFKA_SASL_CONNECT=localhost:9097 cargo test

in another session. Note that Apache Kafka supports a different set of features then redpanda, so we pass other environment variables.

Using a SOCKS5 Proxy

To run the integration test via a SOCKS5 proxy, you need to set the environment variable SOCKS_PROXY. The following command requires a running proxy on the local machine.

bash
$ TEST_INTEGRATION=1 TEST_BROKER_IMPL=kafka KAFKA_CONNECT=0.0.0.0:9011,kafka-1:9021,redpanda-1:9021 SOCKS5_PROXY=localhost:1080 cargo test --features full

The SOCKS5 proxy will automatically be started by the docker compose files. Note that KAFKA_CONNECT was extended by addresses that are reachable via the proxy.

Java Interopt

To test if RSKafka can produce/consume records to/from the official Java client, you need to have Java installed and the TEST_JAVA_INTEROPT=1 environment variable set.

Fuzzing

RSKafka offers fuzz targets for certain protocol parsing steps. To build them make sure you have cargo-fuzz installed. Select one of the following fuzzers:

  • protocol_reader: Selects an API key and API version and then reads message frames and tries to decode the response object. The message frames are read w/o the length marker for more efficient fuzzing.
  • record_batch_body_reader: Reads the inner part of a record batch (w/o the prefix that contains length and CRC) and tries to decode it. In theory this is covered by protocol_reader as well but the length fields and CRC make it hard for the fuzzer to traverse this data structure.

Then run the fuzzer with:

bash
$ cargo +nightly fuzz run protocol_reader
...

Let it running for how long you wish or until it finds a crash:

…

Sadly the backtraces that you might get are not really helpful and you need a debugger to detect the exact source locations:

…

Then create a unit test and fix the bug.

For out-of-memory errors LLDB does not stop automatically. You can however set a breakpoint before starting the execution that hooks right into the place where it is about to exit:

bash
(lldb) b fuzzer::PrintStackTrace()

Benchmarks

Install cargo-criterion, make sure you have some Kafka cluster running, and then you can run all benchmarks with:

bash
$ TEST_INTEGRATION=1 TEST_BROKER_IMPL=kafka KAFKA_CONNECT=localhost:9011 cargo criterion --all-features

If you find a benchmark that is too slow, you can may want to profile it. Get cargo-with, and perf, then run (here for the parallel/rskafka benchmark):

bash
$ TEST_INTEGRATION=1 TEST_BROKER_IMPL=kafka KAFKA_CONNECT=localhost:9011 cargo with 'perf record --call-graph dwarf -- {bin}' -- \
    bench --all-features --bench write_throughput -- \
    --bench --noplot parallel/rskafka

Have a look at the report:

bash
$ perf report

License

Licensed under either of these:

  • Apache License, Version 2.0 (LICENSE-APACHE or )
  • MIT License (LICENSE-MIT or )

Contributing

Unless you explicitly state otherwise, any contribution you intentionally submit for inclusion in the work, as defined in the Apache-2.0 license, shall be dual-licensed as above, without any additional terms or conditions.

[^1]: Kafka's design makes it hard for any client to support the converse, as ultimately each partition is an independent write stream within the broker. However, this crate makes no attempt to mitigate per-partition overheads e.g. by batching writes to multiple partitions in a single ProduceRequest

Issues· 0 open

View all issuesOpen on GitHub

No open issues yet, or sync has not completed.

> Tags

Rust

No comments yet. Be the first to share.

> Details

PublishedAug 1, 2026
UpdatedSep 18, 2026
Category编程语言
PricingOpen source

> Related tools

T
TypeScript
JavaScript 的超集,为前端与全栈提供静态类型
P
Python
通用编程语言,广泛用于 Web、数据与 AI
G
Go
Google 推出的简洁高效系统语言