3.9 KiB
title | authors | creation-date | last-updated |
---|---|---|---|
Adding a New Chaos Fault - Load Testing with k6 | @namkyu1999 | 2023-11-23 | 2023-11-23 |
Adding a New Chaos Fault - Load Testing with k6
- Summary
- Motivation
- Proposal
- Risks and Mitigations
- Upgrade / Downgrade Strategy
- Drawbacks
- Alternatives
- References
Summary
k6 is an open-source load testing tool managed by Grafana Labs. It supports various types of load tests, is easy to use(simply write js script), and is fully open-source. So I want to add a new chaos fault(k6 load test) to Litmus ChaosHub.
Motivation
Litmus ChaosHub has plenty of Chaos Faults. And we already have Pod I/O stress and Node I/O stress. But These do not cover all of Load testing. k6 supports many types of load testing such as smoke testing and stress testing. You can check all types in this url. Plus, this open-source is already mature, with 22k GitHub stars and a strong community with robust use cases. Adding 'k6 load testing' Chaos Fault in ChaosHub can help create a more resilient system.
Goals
- Adding a 'k6 load testing' Chaos Fault to Litmus ChaosHub
- Fixing litmus-go and chaos-charts codes
Non-Goals
- Fixing k6 codes is a non-goal
- Using k6 operator is a non-goal
- Fixing Litmus codes except for litmus-go and chaos-charts is a non-goal
Proposal
Use Cases
Use case 1
In Chaos Studio, Users can select 'k6 load testing' Chaos Fault as part of the Chaos Experiment. They can compose it with other Chaos Faults.
Implementation Details
Here's a Chaos Fault Scenario.
Phase 1 - Add scenario to the litmus-go repository
I will use litmuschaos/go-runner
image. So I am going to add a new case in the litmus-go repository.
Phase 2 - Add a new Chaos Fault to the Litmus ChaosHub
After Phase 1 PR gets merged, I will raise a PR that adds a 'k6 load testing' Chaos Fault to the chaos-charts
repository. When all is done, the user can easily use k6 in Chaos Studio.
Risks and Mitigations
We need to grant proper RBAC permissions to the runner container. Granting override permissions may affect other systems.
Upgrade / Downgrade Strategy
Drawbacks
Alternatives
As I mentioned earlier, we already have Pod I/O stress and Node I/O stress. If we adopt k6, we can test not only in the pod, and node layer but also service, load-balancer, or any layer that has ip.
Locust can be a alternative. it supports python script based load testing. I prefer k6 over locust because k6 supports more load testing types and has cloud-native open-source integrations like Prometheus and Grafana.