Skip to content

🐒 🔥 Datadog Failure Injection System for Kubernetes

License

Notifications You must be signed in to change notification settings

zendesk/chaos-controller-fork

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Oldest Kubernetes version supported: 1.16

⚠️ Kubernetes version 1.20.x is not supported! This Kubernetes issue prevents the controller from running properly on Kubernetes 1.20.0-1.20.4. Earlier versions of Kubernetes as well as 1.20.5 and later are still supported.

Datadog Chaos Controller

💣 Disclaimer 💣

The Chaos Controller allows you to disrupt your Kubernetes infrastructure through various means including but not limited to: bringing down resources you have provisioned and preventing critical data from being transmitted between resources. The use of Chaos Controller on your production system is done at your own discretion and risk.

The Chaos Controller is a Kubernetes controller with which you can inject various systemic failures, at scale, and without caring about the implementation details of your Kubernetes infrastructure. It was created with a specific mindset answering Datadog's internal needs:

  • 🐇 Be fast and operate at scale
    • At Datadog, we are running experiments injecting and cleaning failures to/from thousands of targets within a few minutes.
  • 🚑 Be safe and operate in highly disrupted environments
    • The controller is built to be able to limit the blast radius of failures but also to be able to recover by itself in catastrophic scenarios.
  • 💡 Be smart and operate in various technical environments
    • With Kubernetes, all environments are built differently.
    • Whatever your cluster configuration and implement details choice, the controller is able to inject failures by relying on low-level Linux kernel features such as cgroups, tc or even eBPF.
  • 🪙 Be simple and operate at low cost
    • Most of the time, your Chaos Engineering platform is waiting and doing nothing.
    • We built this project so it uses resources only when it is really doing something:
      • No DaemonSet or any always-running processes on your nodes for injection, no reserved resources when it's not needed.
      • Injection pods are created only when it is needed, killed once experiment is done, and built to be evicted if necessary to free resources.
      • A single long-running pod, the controller, and nothing else!

Getting Started

💡 Read the latest release quick installation guide and the configuration guide to know how to deploy the controller.

Disruptions are built as short-living resources which should be manually created and removed once your experiments are done. They should not be part of any application deployment. The Disruption resource is immutable. Once applied, you can't edit it. If you need to change the disruption definition, you need to delete the existing resource and to re-create it.

Getting started is as simple as creating a Kubernetes resource:

apiVersion: chaos.datadoghq.com/v1beta1
kind: Disruption
metadata:
  name: node-failure
  namespace: chaos-demo # it must be in the same namespace as targeted resources
spec:
  selector: # a label selector used to target resources
    app: demo-curl
  count: 1 # the number of resources to target, can be a percentage if you suffix with "%", e.g. `count: 50%`
  duration: 1h # the amount of time before your disruption automatically terminates itself, for safety
  nodeFailure: # trigger a kernel panic on the target node
    shutdown: false # do not force the node to be kept down

To disrupt your cluster, run kubectl apply -f <disruption_file>.yaml. You can clean up the disruption with kubectl delete -f <disruption_file>.yaml. For your safety, we recommend you get started with the dry-run mode enabled.

📖 The features guide details all the features of the Chaos Controller.

📖 The examples guide contains a list of various disruption files that you can use.

Check out Chaosli if you want some help understanding/creating disruption configurations.

Chaos Scheduling

New feature in 8.0.0

The Chaos Controller has expanded its capabilities by introducing disruption scheduling, enhancing your ability to automate and test system resilience consistently. Instead of manual creation and deletion, use DisruptionCron to regularly disrupt long-lived Kubernetes resources like Deployments and StatefulSets.

Example:

apiVersion: chaos.datadoghq.com/v1beta1
kind: DisruptionCron
metadata:
  name: node-failure
  namespace: chaos-demo
spec:
  schedule: "*/15 * * * *" # every 15 minutes
  targetResource: 
    kind: deployment
    name: demo-curl
  disruptionTemplate:
    count: 1 
    duration: 1h 
    nodeFailure:
      shutdown: false

To schedule disruption in your cluster, run kubectl apply -f <disruption_cron_file>.yaml. To stop, run kubectl delete -f <disruption_cron_file>.yaml.

🔎 Check out DisruptionCron guide for more detailed information on how to schedule disruptions.

Contributing

Chaos Engineering is necessarily different from system to system. We encourage you to try out this tool, and extend it for your own use cases. If you want to run the source code locally to make and test implementation changes, visit the Contributing Doc. By the way, we welcome Pull Requests.

Useful Links

About

🐒 🔥 Datadog Failure Injection System for Kubernetes

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C 79.4%
  • Go 19.7%
  • Python 0.5%
  • Makefile 0.3%
  • Dockerfile 0.1%
  • Shell 0.0%