How ChaosBlade’s Unified Experiment Model Boosts Cloud‑Native Resilience

Introduction

ChaosBlade is an open‑source chaos engineering project originated in 2019 and now part of the CNCF Sandbox. It provides a unified experiment model and a set of executors for Linux, Windows, Docker, Kubernetes and multiple programming languages.

Chaos Experiment Model

Derivation

Fault scenarios are expressed by specifying the affected resource (machine, node, pod, etc.), the component that fails, and the impact. Typical descriptions include a full disk on a host, slow Dubbo service causing upstream latency, or CPU saturation on a Kubernetes node.

Model Definition

Scope : the target machines, clusters or resources where the experiment runs.

Target : the component to inject faults into (e.g., CPU, network, disk, Dubbo, Redis, Pod).

Matcher : rules that select the target; multiple matchers can be combined for complex conditions such as specific RPC service pairs or Redis operations.

Action : the concrete fault to simulate (e.g., disk full, latency, exception, network loss).

Significance

Provides a concise, language‑agnostic representation of fault scenarios.

Facilitates precise description, systematic accumulation and easy extension of scenarios.

ChaosBlade Tool

Supported Scenarios

ChaosBlade covers more than 200 experiment scenarios and over 3,000 parameters across the following domains:

Basic resources: CPU, memory, network, disk, processes, kernel.

Application services: databases, caches, messaging systems, JVM, micro‑services, method‑level fault injection.

Docker containers: container termination, resource exhaustion.

Kubernetes resources: node, pod, container faults.

Cloud resources: Alibaba Cloud ECS failures.

Usage

ChaosBlade is distributed as a binary that can be run in CLI mode or as an HTTP server. Example CLI command to inject packet loss on port 9520: blade create network loss --target 9520 The command returns a UID that can be used to query or destroy the experiment:

blade destroy <UID>

Architecture

The project is modularized per domain. Each domain implements its own executor, for example: chaosblade-exec-os – OS‑level resources. chaosblade-exec-docker – Docker container experiments. chaosblade-operator – Kubernetes CRD‑based experiments. chaosblade-exec-jvm – Java Agent based fault injection. chaosblade-exec-cplus – C++ fault injection via GDB.

The core chaosblade CLI manages the experiment lifecycle, while chaosblade-spec-go defines the experiment model in Go.

ChaosBlade‑Box Platform

Key Features

Hosts multiple open‑source experiment tools (ChaosBlade, LitmusChaos, future Chaos Mesh).

Provides a rich set of scenarios covering resources, multi‑language services and Kubernetes.

Automates tool deployment via Helm and integrates Prometheus for metric collection.

Offers a unified UI for experiment creation, execution and monitoring.

Architecture

The platform automatically deploys the hosted tools, presents a unified experiment model to the UI, and orchestrates experiments across hosts, nodes, pods and containers. Users select resources through a white‑screen interface, launch experiments and view task status and metrics.

Release artifacts and installation instructions are available at https://github.com/chaosblade-io/chaosblade-box/releases.

Usage Workflow

After installing the platform, users configure clusters or hosts, create experiments by selecting dimensions (host, node, pod, container), choose from the hosted scenarios and launch tasks. The task detail page shows experiment status and provides control actions such as pause or destroy.

Future Plans

ChaosBlade will continue to expand scenario coverage, improve stability, and support additional Kubernetes resources and standardized application‑service fault models. ChaosBlade‑Box aims to open‑source core functions of the Alibaba Cloud Fault‑Drill Platform, integrate more third‑party tools, provide scenario recommendations and generate comprehensive experiment reports to close the chaos engineering loop.