ChaosBlade is a chaos engineering toolkit originally built at Alibaba from the lessons of their internal MonkeyKing fault-injection system. It’s distributed as a single blade CLI that can inject a wide catalog of faults into Linux hosts, JVM applications, Docker and containerd containers, Kubernetes clusters, and cloud resources — CPU load, memory pressure, disk fill, network latency and loss, process kill, file corruption, JVM method exceptions and delays, and database query faults among them.
The core is Go, with language-specific agents where needed. For Linux OS-level faults it wraps utilities like tc, stress-ng, and iptables behind a unified grammar. For JVM faults, a Java agent uses bytecode instrumentation to inject exceptions or delays into specific methods without restarting the application. For Kubernetes, a chaosblade-operator exposes ChaosBlade CRDs so you can describe experiments declaratively and target pods via label selectors.
ChaosBlade joined the CNCF sandbox in 2021. In the chaos engineering space it sits next to Chaos Mesh (also CNCF, more Kubernetes-native and with a richer UI), LitmusChaos, and the commercial Gremlin. ChaosBlade’s strongest points are the breadth of OS-level fault types and the JVM bytecode injection, which matches Alibaba’s Java-heavy stack. Outside of JVM-dominated environments, most new adopters pick Chaos Mesh or Litmus instead.