Chaos Experiment
Chaos Experiment is a set of different operations coupled together to achieve desired chaos impact on a Kubernetes Cluster.
It is useful in automating a series of pre-conditioning steps or action which is necessary to be performed before triggering the chaos injection.
A Chaos Experiment can also be used to perform different operations parallelly to achieve a desired chaos impact.
note
With the latest release of LitmusChaos 3.0.0:
Prerequisites​
The following should be required before creating a Chaos Experiment:
How do we define and execute a Chaos Experiment ?​
LitmusChaos leverages the popular GitOps tool Argo to achieve this goal. Argo enables the creation of different chaos experiments together in form of chaos experiments which are extremely simple and efficient to use.
With the help of ChaosCenter, chaos experiments with different types of faults can be created. In a Chaos Experiment, the faults can be set to execute in parallel to each other and the user can tune the chaos experiment by adding additional steps to simulate a desired fault that might occur in the production stage.
Life Cycle of a Chaos Experiment​
Here is a sample pod-delete chaos experiment from ChaosCenter.
apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
name: custom-chaos-workflow-1627980541
namespace: litmus
labels:
subject: custom-chaos-workflow_litmus
spec:
arguments:
parameters:
- name: adminModeNamespace
value: litmus
entrypoint: custom-chaos
securityContext:
runAsNonRoot: true
runAsUser: 1000
serviceAccountName: argo-chaos
templates:
- name: custom-chaos
steps:
- - name: install-chaos-experiments
template: install-chaos-experiments
- - name: pod-delete
template: pod-delete
- - name: revert-chaos
template: revert-chaos
- name: install-chaos-experiments
inputs:
artifacts:
- name: pod-delete
path: /tmp/pod-delete.yaml
raw:
data: >
apiVersion: litmuschaos.io/v1alpha1
description:
message: |
Deletes a pod belonging to a deployment/statefulset/daemonset
kind: ChaosExperiment
metadata:
name: pod-delete
labels:
name: pod-delete
app.kubernetes.io/part-of: litmus
app.kubernetes.io/component: chaosexperiment
app.kubernetes.io/version: 3.0.0
spec:
definition:
scope: Namespaced
permissions:
- apiGroups:
- ""
- apps
- apps.openshift.io
- argoproj.io
- batch
- litmuschaos.io
resources:
- deployments
- jobs
- pods
- pods/log
- replicationcontrollers
- deployments
- statefulsets
- daemonsets
- replicasets
- deploymentconfigs
- rollouts
- pods/exec
- events
- chaosengines
- chaosexperiments
- chaosresults
verbs:
- create
- list
- get
- patch
- update
- delete
- deletecollection
image: litmuschaos/go-runner:3.0.0
imagePullPolicy: Always
args:
- -c
- ./experiments -name pod-delete
command:
- /bin/bash
env:
- name: TOTAL_CHAOS_DURATION
value: "15"
- name: RAMP_TIME
value: ""
- name: FORCE
value: "true"
- name: CHAOS_INTERVAL
value: "5"
- name: PODS_AFFECTED_PERC
value: ""
- name: LIB
value: litmus
- name: TARGET_PODS
value: ""
- name: SEQUENCE
value: parallel
labels:
name: pod-delete
app.kubernetes.io/part-of: litmus
app.kubernetes.io/component: experiment-job
app.kubernetes.io/version: 3.0.0
container:
args:
- kubectl apply -f /tmp/pod-delete.yaml -n
{{workflow.parameters.adminModeNamespace}} | sleep 30
command:
- sh
- -c
image: litmuschaos/k8s:latest
- name: pod-delete
inputs:
artifacts:
- name: pod-delete
path: /tmp/chaosengine-pod-delete.yaml
raw:
data: |
apiVersion: litmuschaos.io/v1alpha1
kind: ChaosEngine
metadata:
namespace: "{{workflow.parameters.adminModeNamespace}}"
generateName: pod-delete
labels:
instance_id: 86a4f130-d99b-4e91-b34b-8f9eee22cb63
spec:
appinfo:
appns: default
applabel: app=nginx
appkind: deployment
jobCleanUpPolicy: retain
engineState: active
chaosServiceAccount: litmus-admin
experiments:
- name: pod-delete
spec:
components:
env:
- name: TOTAL_CHAOS_DURATION
value: "30"
- name: CHAOS_INTERVAL
value: "10"
- name: FORCE
value: "false"
- name: PODS_AFFECTED_PERC
value: ""
container:
args:
- -file=/tmp/chaosengine-pod-delete.yaml
- -saveName=/tmp/engine-name
image: litmuschaos/litmus-checker:latest
- name: revert-chaos
container:
image: litmuschaos/k8s:latest
command:
- sh
- -c
args:
- "kubectl delete chaosengine -l 'instance_id in
(86a4f130-d99b-4e91-b34b-8f9eee22cb63, )' -n
{{workflow.parameters.adminModeNamespace}} "
podGC:
strategy: OnWorkflowCompletion
The structure of a chaos experiment is similar to that of a Kubernetes Object. It consists of mandatory fields like apiVersion, kind, metadata, spec.
The spec in a Chaos Experiment is where the different steps are mentioned and the overall life cycle of the chaos experiment is described.
We can see different templates are present in the spec of a chaos experiment.
templates:
- name: custom-chaos
steps:
- - name: install-chaos-experiments
template: install-chaos-experiments
- - name: pod-delete
template: pod-delete
- - name: revert-chaos
template: revert-chaos
Here in this template, we can see different steps are present. These include installing the chaos faults, executing the chaos engine of the faults, and at the end we have the revert chaos step which deletes/removes the resources that were created as part of the chaos experiment.
Some additional checks can be added with the faults in the form of probes. These probes are defined in the ChaosEngines of the faults and are updated when the fault execution takes place.
The overall chaos experiment result can be viewed with the ChaosResult CRD which contains the verdict and the probeSuccessPercentage (a ratio of successful checks v/s total probes).
What is a run?​
A chaos experiment run can be defined as a single/one-time execution of the chaos experiment. There can be multiple runs of a single chaos experiment. If the chaos experiment consists of a cron syntax, it will run periodically according to the cron provided in the chaos experiment.
What is Resilience Score?​
Resiliency score is an overall measure of the resiliency of a system for a given chaos experiment, which is obtained upon executing the constituent experiment faults on that system.
While creating a chaos experiment, certain weights are assigned to all the faults present in the chaos experiment. These weights signify the priority/importance of the fault. The higher the weight, the more significant the fault is.
In ChaosCenter, the weight priority is generally divided into three sections:
- 0-3: Low Priority
- 4-6: Medium Priority
- 7-10: High Priority
Once a weight has been assigned to the fault, we look for the Probe Success Percentage for that fault itself (Post Chaos) and calculate the total resilience result for that fault as a multiplication of the weight given and the probe success percentage returned after the Chaos Run.
Total Resilience for one single fault = (Weight Given to that fault * Probe Success Percentage)
Overall Resilience Score = Total Test Result / Sum of the assigned weights of the faults
What is a Cron Chaos Experiment?​
Cron Chaos Experiment is a type of chaos experiment that runs on a pre-defined schedule. It consists of a mandatory field spec.schedule. A cron syntax is provided in this field at which the chaos experiment execution takes
place.
Here's a sample Cron Chaos Experiment for Podtato-Head application:
apiVersion: argoproj.io/v1alpha1
kind: CronWorkflow
metadata:
name: podtato-head-1628058291
namespace: litmus
labels:
subject: podtato-head_litmus
spec:
schedule: 10 0-23 * * *
concurrencyPolicy: Forbid
startingDeadlineSeconds: 0
workflowSpec:
entrypoint: argowf-chaos
serviceAccountName: argo-chaos
securityContext:
runAsUser: 1000
runAsNonRoot: true
arguments:
parameters:
- name: adminModeNamespace
value: litmus
templates:
- name: argowf-chaos
steps:
- - name: install-application
template: install-application
- - name: install-chaos-experiments
template: install-chaos-experiments
- - name: pod-delete
template: pod-delete
- - name: revert-chaos
template: revert-chaos
- name: delete-application
template: delete-application
- name: install-application
container:
image: litmuschaos/litmus-app-deployer:latest
args:
- -namespace={{workflow.parameters.adminModeNamespace}}
- -typeName=resilient
- -operation=apply
- -timeout=400
- -app=podtato-head
- -scope=namespace
- name: install-chaos-experiments
container:
image: litmuschaos/k8s:latest
command:
- sh
- -c
args:
- kubectl apply -f
https://hub.litmuschaos.io/api/chaos/1.13.7?file=charts/generic/experiments.yaml
-n {{workflow.parameters.adminModeNamespace}} ; sleep 30
- name: pod-delete
inputs:
artifacts:
- name: pod-delete
path: /tmp/chaosengine.yaml
raw:
data: >
apiVersion: litmuschaos.io/v1alpha1
kind: ChaosEngine
metadata:
namespace: "{{workflow.parameters.adminModeNamespace}}"
labels:
instance_id: 1b7ec920-75f9-4398-b4c3-9c3a5d7fd5c2
generateName: podtato-main-pod-delete-chaos
spec:
appinfo:
appns: "{{workflow.parameters.adminModeNamespace}}"
applabel: name=podtato-main
appkind: deployment
engineState: active
chaosServiceAccount: litmus-admin
jobCleanUpPolicy: retain
components:
runner:
imagePullPolicy: Always
experiments:
- name: pod-delete
spec:
probe:
- name: check-podtato-main-access-url
type: httpProbe
httpProbe/inputs:
url: http://podtato-main.{{workflow.parameters.adminModeNamespace}}.svc.cluster.local:9000
insecureSkipVerify: false
method:
get:
criteria: ==
responseCode: "200"
mode: Continuous
runProperties:
probeTimeout: 1
interval: 1
retry: 1
components:
env:
- name: TOTAL_CHAOS_DURATION
value: "30"
- name: CHAOS_INTERVAL
value: "10"
- name: FORCE
value: "false"
container:
image: litmuschaos/litmus-checker:latest
args:
- -file=/tmp/chaosengine.yaml
- -saveName=/tmp/engine-name
- name: delete-application
container:
image: litmuschaos/litmus-app-deployer:latest
args:
- -namespace={{workflow.parameters.adminModeNamespace}}
- -typeName=resilient
- -operation=delete
- -app=podtato-head
- name: revert-chaos
container:
image: litmuschaos/k8s:latest
command:
- sh
- -c
args:
- "kubectl delete chaosengine -l 'instance_id in
(1b7ec920-75f9-4398-b4c3-9c3a5d7fd5c2, )' -n
{{workflow.parameters.adminModeNamespace}} "
timezone: Asia/Calcutta
In the above chaos experiment, we can see the cron syntax at spec.schedule is
spec:
schedule: 10 0-23 * * *
This means the chaos experiment will be executed at the 10th minute of every hour.
A chaos experiment can be changed into Cron Chaos Experiment from the ChaosCenter.
While scheduling a chaos experiment, in the Schedule step, there are few options as part of Recurring Schedules. These include:
- Every hour
- Every Day
- Every Week
- Every Month
Summary​
A chaos experiment is a combination of different steps combined together to perform a specific chaos use-case on a Kubernetes system. These steps can include installing fault steps, ChaosEngine CR for target selection, revert-chaos steps, etc. Chaos Experiments can be scheduled for a later time with the help of Cron Chaos Experiments. These chaos experiments consist of a cron syntax that is used for scheduling a chaos experiment. Once the chaos experiment execution is completed, the resiliency of the targeted application is calculated. Several weights are assigned to different faults in the chaos experiment. These weights are used along with the ProbeSuccessPercentage to find out the resiliency score.