Skip to content

Run Chaos Toolkit with a Google Cloud Platform Cloud Run Job

Google Cloud Platform Cloud Run Job is an ideal candidate when you need to run a Chaos Toolkit experiment once or recurringly.


This page describe the basics for running an experiment as a Cloud Run Job. It should provide the skeleton for your own requirements.

In a nutshell, you will need:

  • a container image with the Chaos Toolkit and the right extensions
  • an experiment file. For the purpose of this documentation, we will host the experiment in a Google Cloud Storage bucket. It can be stored anywhere else that can be access over HTTP otherwise

Create a Cloud Run Job

Below is an example of a Cloud Run Job manifest to run an experiment:

kind: Job
  name: chaostoolkit-experiment
  annotations: chaostoolkit BETA # (1)!
      annotations: gen2
          serviceAccountName: chaostoolkit-cloud-run-job
          maxRetries: 0 # (2)!
          timeoutSeconds: 3600 # (3)!
            - image: chaostoolkit/chaostoolkit:full # (8)!
                - --verbose
                - run
                - /home/svc/experiment/experiment.json
                  memory: 512Mi
                  cpu: "1"
              - mountPath: /home/svc/experiment  # (4)!
                name: experiments
                subPath: experiment.json  # (5)!
          - name: experiments
              driver:  # (6)!
              readOnly: true
                bucketName: my_bucket  # (7)!
  1. Required only to mount the Cloud Storage as a volume
  2. A good idea to not retry automatically to run the experiment
  3. Not strictly necessary but a good habit to control your bills
  4. This is the directory we mount into the container to hold our experiment from the volume
  5. The name of the Cloud Storage object in our bucket
  6. Specific to Cloud Storage
  7. Name of the bucket containing the experiment
  8. Let’s use a base image that contains a list of extensions

To create the job, we then run the following commands. First we create the necessary service account and bind the appropriate role to mount the Cloud Storage to it:

export PROJECT_ID=  # (1)!
export REGION=

gcloud config set run/region $REGION  # (2)!

gcloud iam service-accounts create chaostoolkit-cloud-run-job \  # (3)!
    --description="Runs Chaos Toolkit Experiments" \
    --display-name="Dedicated service account to run Chaos Toolkit experiments"

gcloud projects add-iam-policy-binding ${PROJECT_ID} \  # (4)!
    --member="serviceAccount:chaostoolkit-cloud-run-job@${PROJECT_ID}" \
    --role="roles/storage.objectViewer" \
  1. Set the project identifier to your project
  2. Set the correct region where to create and run the job
  3. This is the service account associated with the job
  4. We only need the "roles/storage.objectViewer" to mount the cloud storage as a volume

We now need a space to store our experiment file.

export REGION=
export BUCKET=my_bucket

gcloud storage buckets create gs://$BUCKET --location $REGION  # (1)!

gcloud storage cp ./experiment.json gs://$BUCKET  # (2)!
  1. The bucket name is the one we set in the job configuration above
  2. Copy the experiment.json file at the root of the bucket

Next, we can can now create the Cloud Run job:

gcloud run jobs replace job.yaml

You can view your job as well:

gcloud run jobs describe chaostoolkit-experiment

Run the Experiment Once

Once you have created a job, you can run it at will as follows:

gcloud run jobs execute --wait chaostoolkit-experiment

This will run and wait for the job to complete and will output the URL of the execution.

Schedule the Experiment Repeatedly

To schedule the job to run recurringly, you will need another Google Cloud service named the Scheduler.

Here is an example on how to schedule the experiment to run every Monday at 9am:

export PROJECT_ID=  # (1)!
export REGION= # (2)!

gcloud iam service-accounts create chaostoolkit-job-sched \
    --description="Schedules Chaos Toolkit Experiments" \
    --display-name="Dedicated service account to schedule Chaos Toolkit experiments"

gcloud projects add-iam-policy-binding ${PROJECT_ID} \
    --member="serviceAccount:chaostoolkit-job-sched@${PROJECT_ID}" \
    --role="roles/run.invoker" \  # (3)!

gcloud scheduler jobs create http my-experiment-schedule \
  --location ${REGION} \
  --http-method POST \
  --schedule="0 9 * * 1" \
  --time-zone "Etc/UTC" \  # (4)!
  --uri="https://${REGION}${PROJECT_ID}/jobs/chaostoolkit-experiment:run" \
  --oauth-service-account-email chaostoolkit-job-sched@${PROJECT_ID}
  1. Set the project identifier to your project
  2. Set the appropriate region where to run the scheduler from. The scheduler and the job do not need to reside in the same region
  3. This role is required so the service account used to trigger the job is allowed to do it
  4. A good idea to set the correct timezone

You can test the schedule without waiting for the first instance:

gcloud scheduler jobs run --location=${REGION} my-experiment-schedule