Goal:

Load testing the new tenant creation scenario on weavaite cluster deployed on a multi node cluster on k8s.

Framework:

The framework uses k6 load testing tool to simulate the tenant creation scenario

Install the k6 performance test tool.

Install dependencies using:

yarn install

in the terminal (you need to have yarn installed on your machine).

Build the tests using the following command:

yarn build

Run the test using the following command:

yarn test

Threshold Configuration

The threshold configuration is captured in config.json file. Please see k6 Options documentation on various K6 options that can be used to configure the tests. E.g constant-vus, ramping-vus etc

Configuration:

The framework contains the config.json file where weavaite cluster's host url, weaviate api Key, openai api keys are configured. If there are multiple cluster then mention weaviates's instance hosturl, api key for every cluster. e.g

` { "Cluster": [ { "host": "", "apiKey": "", "X-OpenAI-Api-Key": "" }, { "host": "", "apiKey": "", "X-OpenAI-Api-Key": "" }, { "host": "", "apiKey": "", "X-OpenAI-Api-Key": "" } ] }

`

Metrics:

The key metrics include: - http_req_duration, the end to end time of all the requests (that is the total layency) - http_req_failed, the total number of failed requests - iterations, the total number of iterations - Peak RPS , the peak request per sec - http_req , Total http request made - response time, Response time to process the request - The test results show aggregated statistical values including, Median and average values , Minimum and Maximum values , p90, p95 and p99 values

Dashboard:

The test execution summary is visalized in grafana for deeper analysis. This gives you a granular time-series data, which has metrics and timestamps for evey point of the test. https://rthiiyer.grafana.net/a/k6-app/tests/830424

Improvements:

- Create a docker compose file to deploy the laod test framework on multinode k8 cluster
- Run the test on cloud instead of local execution
- Schedule to run the tests on regualr basis to monitor the performance of the system.
- Add more scenarios to the framework including loading of dataset (large, medium and small) in JSON, CSV format (currently supported)