Getting Started

Verified against the v1.8.0 release line. Commands assume a Linux host with tar and curl, and a shell where you can run blade with the privileges the chosen fault needs (CPU load needs none special; some network or disk faults need root).

Prerequisites

• A Linux x86-64 host (the release ships per-OS, per-arch toolkits).
• curl and tar to fetch and unpack a release.
• For Kubernetes faults: a reachable cluster, kubectl, and helm.

Install

Download a release toolkit from the GitHub releases page and unpack it. The tarball contains the blade binary plus the bin/ executors and the versioned YAML specs.

curl -sSL -o chaosblade.tar.gz \
  https://github.com/chaosblade-io/chaosblade/releases/download/v1.8.0/chaosblade-1.8.0-linux-amd64.tar.gz
tar -xzf chaosblade.tar.gz
cd chaosblade-1.8.0

A first working setup

The shortest end-to-end run is a CPU load experiment: create it, confirm it, then destroy it. The uid printed by create is what you pass to status and destroy.

1. Inject 60% CPU load. The command prints a JSON response whose result field is the experiment uid.

   ./blade create cpu load --cpu-percent 60

2. Query the experiment by the uid returned above.

   ./blade status <experiment-uid>

3. Stop the experiment and recover the host.

   ./blade destroy <experiment-uid>

To avoid a forgotten experiment, add --timeout so blade schedules its own destroy. This run recovers automatically after 30 seconds:

./blade create cpu load --cpu-percent 60 --timeout 30

For Kubernetes, install the operator with Helm, then create a pod fault. The operator exposes experiments as a Custom Resource Definition (CRD).

helm install chaosblade-operator chaosblade-operator-1.8.0.tgz \
  --namespace chaosblade --create-namespace
./blade create k8s pod-cpu fullload --cpu-percent 80 \
  --kubeconfig ~/.kube/config --names <pod-name> --namespace default

Verify it works

create returns a JSON object with "code": 200 and a "result" holding the uid on success. Run ./blade status <experiment-uid> and check that the experiment shows as Success. While a CPU load experiment is active, top on the host should show elevated CPU usage; after destroy (or after the --timeout elapses) it returns to baseline. State is recorded in the local chaosblade.dat SQLite file next to the binary.

Where to go next

• Official docs cover the full scenario catalogue (network, disk, process, JVM, container): https://chaosblade.io/en/docs/
• The Kubernetes operator and its CRD reference: https://github.com/chaosblade-io/chaosblade-operator
• The chaosblade-box platform adds a UI, Prometheus integration, and multi-cluster management: https://chaosblade.io/en/blog/2022/06/24/ChaosBlade-Box-a-New-Version-of-the-Chaos-Engineering-Platform-Has-Released/