Key values

The following key values of the Chaos Toolkit reflect the mindset the community has when engineering the Chaos Toolkit project.


The Chaos Toolkit aims at being a simple piece of technology both from a user and developer perspective.

To achieve simplicity, the Chaos Toolkit comes as a command line interface driven by a description file. As a user, this means no code and no need to learn a programming language. As a developer, this reduces the functional surface area to consider.


The Chaos Toolkit does not wish to be a monolith and strives to be extended to fully reach its goal through community driven efforts.

By using a description file, the implementation is not prescribed by the Chaos Toolkit project. Although we fully expect the community to eventually settle on certain implementations of probes and actions.


We believe code readbility is a factor for positive maintenance and evolutivity.

Readable code never goes out of fashion. As the code of the Chaos Toolkit is mostly written in Python, best practices such as defined in PEP8.


Although not strictly speaking referring to the technological aspect of the project, having diversity in the community will contribute to a better project overall.

Core Projects

The Chaos Toolkit is made of several projects that work together to provide its service.


The chaostoolkit project is the command-line interface (CLI), in other words the command executed by users to run their experiments.

That project tries to remain as shallow as possible, only providing the user interface commands by gluing other projects together.

This project is implemented in Python 3.


The chaostoolkit-lib project is the core library which implements the core concepts of the Chaos Toolkit.

This project is implemented in Python 3.


The chaostoolkit-documehtation is the documentation source and renderer of the Chaos Toolkit. Namely, that project generates the website you are currently reading.

This project is implemented in Python 3 by generating HTML from Markdown documents.

Extension Projects

In addition to the core projects, the Chaos Toolkit manages some extension projects which provide probes and/or actions for experiments.


The chaostoolkit-kubernetes project implements probes and actions for experiments targetting a Kubernetes cluster. Those activities are implemented as Python functions.

This project is implemented in Python 3.


The chaostoolkit-gremlin project implements actions for experiments exploring resource failures (CPU, Memory, Network…) in their system through the Gremlin, Inc. services. Those activities are implemented as Python functions talking to the Gremlin API.

This project is implemented in Python 3.


The chaostoolkit-prometheus project implements probes to fetch information from your system through Prometheus. Those probes are implemented as Python functions talking to the Prometheus API.

This project is implemented in Python 3.

Technical Choices

Python 3

The Chaos Toolkit is implemented in Python 3. A high-level language with a long successfuly story for writing great software. It’s a common choice for tooling purpose.

The language supports readbility well and has a large ecosystem of libraries. It is also well-spread and easy to install. The choice to not support Python 2 is a look at Python’s present and future.

The choice for a dynamic language was also motivated because the Chaos Toolkit manipulates a lot of strings and that task is made straightforward with Python.

Although Python cannot generate (well, not easily) standalone binaries like golang would. We do not believe this will harm the project and hope that package managers will eventually provide native installers.


Well, this project is not truly a functional piece of code but the code relies as little as possible on stateful constructions as provided by classes.

Mutable data structures are used but mostly created and returned from functions rather than modified.

Generally speaking, the project draws inspirations from certain ideas of functional paradigms but does not enforce them strictly. One notable area where the code strays away from these principles is the use of exceptions rather than returning error codes. This may change if the community expresses such an intention.


The experiment description and structure is encoded using JSON. The choice for JSON over YAML is because it leaves less room for ambiguity and is marginally less readable for a structure with a shallow depth like Chaos Toolkit experiments.