Back to Devoured - August 27, 2026
DEVOURED
Khaos: Kafka Load Testing & Chaos Engineering (GitHub Repo)

Khaos: Kafka Load Testing & Chaos Engineering (GitHub Repo)

Data devopsinfrastructure GitHub
Khaos is a new Go-based tool for simulating Kafka broker failures and consumer lag without requiring changes to your application code.
What: Khaos, now rewritten in Go, allows platform teams to define and run Kafka failure scenarios like hot partitions and rebalance storms via simple YAML files.
Why it matters: Reliable messaging systems often fail during edge cases; simulating these before production deployments is essential for building resilient event-driven architectures.
Takeaway: Install the latest version using `go install github.com/aleksandarskrbic/khaos/cmd/khaos@latest` to test your Kafka clusters against bundled failure scenarios.
Deep dive
  • Supports generating structured records in JSON, Avro, and Protobuf.
  • Manages local Kafka clusters via Docker Compose or targets existing remote clusters.
  • Uses YAML for scenario definitions, avoiding the need for client libraries or instrumentation.
  • Designed for headless CI execution, ensuring identical behavior between local TUI and automated testing.
  • Uses the pure-Go franz-go client to maintain a small binary footprint without CGO dependencies.
Decoder
  • Hot partition: A Kafka partition receiving significantly more traffic than others, leading to uneven load and performance bottlenecks.
  • Rebalance storm: A repeated cycle of Kafka consumer group rebalancing that can lead to significant processing downtime.
Original article

Khaos: Kafka Load Testing & Chaos Engineering

Khaos is an open-source Kafka traffic generator, load-testing tool, and chaos engineering CLI for reproducing realistic Kafka workloads and failure scenarios (consumer lag, hot partitions, rebalances, and broker failures) on demand, instead of waiting for production to find them.

Documentation · Quick Start · Scenario Reference

What it does

  • Generate realistic Kafka test data: structured, faker-backed records in JSON, Avro, or Protobuf.
  • Simulate producer and consumer traffic: configurable throughput, key distributions, and consumer group topology.
  • Load test Kafka clusters and the applications that consume from them, including Kafka Streams and Flink jobs.
  • Reproduce failure conditions on purpose: consumer lag, hot partitions, rebalances, and broker failures, scheduled on a timeline.

Scenarios are plain YAML. No code, no client library, no instrumentation in the system under test.

Quick start

go install github.com/aleksandarskrbic/khaos/cmd/khaos@latest

khaos list                          # see the bundled scenarios
khaos run traffic/high-throughput   # auto-starts a local 3-broker Kafka cluster

khaos run manages its own local Kafka cluster via Docker Compose. To target a cluster you already have, including managed clusters needing SASL/SSL, use khaos simulate instead. See the Quick Start guide and Installation for release binaries and Docker.

A few scenarios

khaos run traffic/hot-partition       # skewed key distribution overloads one partition
khaos run traffic/consumer-lag        # producer rate outpaces slow consumers
khaos run chaos/broker-chaos          # brokers stop and restart while traffic keeps flowing
khaos run chaos/rebalance-storm       # a consumer group rebalances repeatedly

khaos validate checks a scenario file's structure without running it: the same command Khaos's own CI runs against every bundled scenario. See the Scenarios and Guides sections of the docs for what each one actually configures and why.

Documentation

Full documentation, including the CLI reference, the scenario YAML schema, and guides for load testing, data generation, and each failure scenario, lives at getkhaos.dev/docs:

Architecture

Khaos is a single Go binary. The scenario engine is independent of any user interface: it exposes one read method, Snapshot(), and the terminal UI, the headless log loop, and the final summary table all poll it. Nothing in the engine knows about terminals, so a headless run in CI behaves identically to an interactive one, and a stalled UI can't stall a run.

Kafka access is franz-go, a pure-Go client, which is what makes CGO_ENABLED=0, cross-compilation, go install, and a distroless/static image all work without a C toolchain. See the Concepts page for the full picture.

Rewritten in Go

Khaos was rewritten from Python to Go: a single static binary, no runtime dependencies, and a pure Go Kafka client that cross-compiles cleanly. See the release notes for details.

Contributing

Issues and pull requests are welcome. If Khaos is useful to you, a star helps others find it.

License

Apache 2.0.

Explore all 59 articles in today's digest

Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi