Getting started
Table of contents
Local installation
PowerfulSeal is available to install through pip.
Prerequisites
Python
You’re going to need python 3.7+.
python --version
Python 3.7.6
Setting up a virtualenv
Though optional, it is a good practice to use a virtualenv. You can create and activate one like this:
python -m virtualenv env
source env/bin/activate
Installing through pip
Installing from pip is easy:
pip install powerfulseal
Starting the Seal
To see the syntax of the commands, you can always run:
powerfulseal --help
If you have a kubeconfig file ready (for example in ~/.kube/config) and working with Kubernetes, you can start the seal on defaults:
powerfulseal interactive
This will start a command line, interactive CLI, just like the following. You can type commands inside of it.
(seal) $
Try listing pods from the kube-system namespace:
(seal) $ pods kube-system
For help, just type help. For more information about the modes, see our docs on modes.
Running from Docker
Download docker image
For each release a docker image is built and published to the docker hub.
docker pull powerfulseal/powerfulseal:3.1.1
Run docker image
You can use the docker image in a similar fashion to running locally. You will just need to pass it your kubeconfig.
Below is an example of using the -v flag to inject your local kubeconfig to the image (-v ~/.kube:/root/.kube)
docker run -it \
-v ~/.kube:/root/.kube \
powerfulseal/powerfulseal:3.1.1 \
interactive
To see how to use other modes, see our docs on modes
Hello world!
Write the following hello world policy to policy.yml:
scenarios:
- name: Hello chaos!
description: >
Verifies that after a pod is killed,
it's succesfully rescheduled after 30 seconds.
steps:
# kill a kube-system pod
- podAction:
matches:
- namespace: kube-system
filters:
- randomSample:
size: 1
actions:
- kill:
probability: 1
- wait:
seconds: 30
# make sure all pods are running in the namespace
- podAction:
matches:
- namespace: kube-system
actions:
- checkPodState:
state: Running
You can then run the hello world policy in autonomus mode (the most flexible and powerful mode). Run the following command:
powerfulseal autonomous --policy-file ./policy.yaml
You will see an output similar to the following one. Note, that the Seal killed a pod, and then verified that it was successfuly restarted. By default, it will continue to run this experiment over and over, so feel free to press Ctrl-C to kill it.
$ powerfulseal autonomous --policy-file stuff/policy1.yml
2020-06-05 12:42:03 INFO __main__ verbosity: None; log level: INFO; handler level: INFO
2020-06-05 12:42:03 INFO __main__ Creating kubernetes client with config /Users/chaos/.kube/config (path found for backwards compatibility)
2020-06-05 12:42:03 INFO k8s_client Initializing with config: /Users/chaos/.kube/config
2020-06-05 12:42:03 INFO __main__ No cloud driver - some functionality disabled
2020-06-05 12:42:04 INFO __main__ Using stdout metrics collector
2020-06-05 12:42:04 INFO __main__ Starting the UI server (0.0.0.0:8000)
2020-06-05 12:42:04 INFO __main__ STARTING AUTONOMOUS MODE
2020-06-05 12:42:04 INFO scenario.Hello chaos! Starting scenario 'Hello chaos!' (3 steps)
2020-06-05 12:42:04 INFO action_nodes_pods.Hello chaos! Matching 'namespace' {'namespace': 'kube-system'}
2020-06-05 12:42:04 INFO action_nodes_pods.Hello chaos! Matched 37 pods in namespace kube-system
2020-06-05 12:42:04 INFO action_nodes_pods.Hello chaos! Filtered set length: 1
2020-06-05 12:42:04 INFO action_nodes_pods.Hello chaos! Pod killed: [pod #32 name=kube-state-metrics-7b4944dfbb-zrlrz namespace=kube-system containers=1 ...]
2020-06-05 12:42:04 INFO scenario.Hello chaos! Sleeping for 30 seconds
2020-06-05 12:42:34 INFO action_nodes_pods.Hello chaos! Matching 'namespace' {'namespace': 'kube-system'}
2020-06-05 12:42:35 INFO action_nodes_pods.Hello chaos! Matched 37 pods in namespace kube-system
2020-06-05 12:42:35 INFO action_nodes_pods.Hello chaos! Filtered set length: 37
2020-06-05 12:42:35 INFO scenario.Hello chaos! Scenario finished
We just verified experimentally, that the Kubernetes cluster was able to successfully restart a pod within 30 seconds. Now, what would happen if you changed the wait to 1 second?
...
- wait:
seconds: 1
Most likely this:
...
2020-06-05 12:53:26 INFO action_nodes_pods.Hello chaos! Pod killed: [pod #14 name=kube-apiserver-metrics-pkh6n namespace=kube-system containers=1 ...]
2020-06-05 12:53:26 INFO scenario.Hello chaos! Sleeping for 1 seconds
2020-06-05 12:53:27 INFO action_nodes_pods.Hello chaos! Matching 'namespace' {'namespace': 'kube-system'}
2020-06-05 12:53:27 INFO action_nodes_pods.Hello chaos! Matched 37 pods in namespace kube-system
2020-06-05 12:53:27 INFO action_nodes_pods.Hello chaos! Initial set length: 37
2020-06-05 12:53:27 INFO action_nodes_pods.Hello chaos! Filtered set length: 37
2020-06-05 12:53:27 ERROR action_nodes_pods.Hello chaos! Expected pod in state 'Running', got 'ContainerCreating' ([pod #14 name=kube-apiserver-metrics-f98lw namespace=kube-system containers=1 ...])
2020-06-05 12:53:27 WARNING scenario.Hello chaos! Step returned failure {'podAction': {'matches': [{'namespace': 'kube-system'}], 'actions': [{'checkPodState': {'state': 'Running'}}]}}. Finishing scenario early
2020-06-05 12:53:27 ERROR policy_runner Exiting early
2020-06-05 12:53:27 ERROR __main__ Policy runner finishes with an error
OK, so that’s a taste of what powerfulseal policies are capable of. Learn more about policies here.
Modes of operation
Powerfulseal supports multiple modes of operation.