Chaos Mesh Documentation

What Chaos Mesh is

Chaos engineering means breaking things on purpose, in a controlled way, to learn how a system behaves before a real failure teaches you. Chaos Mesh is an open source platform that does this inside Kubernetes. You install it into a cluster with Helm, then describe an experiment as a Kubernetes custom resource: which pods to target, which fault to inject and for how long. It joined the CNCF in 2020 and moved to the incubating level in February 2022.

What the docs cover

• Getting started: installing with Helm or a script, running it on kind or minikube for practice, and setting user permissions so experiments cannot be run by anyone.
• Fault types: pod faults (pod failure, pod kill, container kill), network faults (delay, loss, partition, bandwidth), stress on CPU and memory, file I/O faults, DNS faults, time skew, JVM application faults, Linux kernel faults, HTTP faults and block device faults, plus faults against AWS, Azure and GCP resources.
• Chaosd: a companion tool for injecting faults into physical machines and virtual machines outside Kubernetes.
• Workflows and schedules: chaining experiments in series or parallel, and running them on a schedule.
• Chaos Dashboard: a web UI for creating, watching and archiving experiments.

Is it worth learning?

If you run services on Kubernetes, yes. Reading about timeouts and retries is one thing; watching your checkout service when its database suddenly adds 500 milliseconds of latency is another. Chaos Mesh makes that experiment a few lines of YAML. It is a good way to test the claims in your own design docs and to rehearse incident response before a real incident.

Prerequisites and time

You need working Kubernetes basics: pods, deployments, services, namespaces and kubectl. A local cluster is enough to practise. Installing and running a first pod kill and network delay takes an afternoon. Designing useful experiments, with a hypothesis and a measured steady state, takes longer and is the real skill.

Free or paid

Chaos Mesh and its documentation are free and open source under the Apache 2.0 licence. There is no paid tier to unlock features.

What you can do after it

• Write a hypothesis such as "if one replica dies, error rate stays under our SLO" and test it.
• Inject latency and packet loss between services to check timeouts and circuit breakers.
• Build a scheduled workflow that runs small experiments continuously in a staging cluster.
• Explain the blast radius of an experiment and how to stop it quickly.

Common questions

Is it safe to run in production? Only with care: start in staging, limit targets with selectors and namespaces, restrict who can create experiments, and know how to abort. The docs cover permissions for this reason.

Chaos Mesh or LitmusChaos? Both are CNCF projects for Kubernetes chaos. Chaos Mesh is known for its broad fault types and simple YAML. Either teaches the same discipline.

Best for: platform engineers and SREs who want evidence that their resilience patterns actually work.