As a part of this [[ https://phabricator.wikimedia.org/T340712 | ticket ]], we tested liftwing API from a home connection. The datacenter hosting liftwing API server has a peering connection with aws. So, we are supposed to see better response times when the test script is run from aws.
In this ticket, by re-running the test from aws, we would like to verify the following for liftwing revertrisk-language-agnostic endpoint:
- we are able to send upto 250 K requests in a 1 hour duration. Beyond this, requests should error out.
- for large percentage of requests, the response time is <= 500 ms.
**Notes:**
1. Use a user agent as prescribed by the ML team for the requests.
2. During the span of an hour, 250K requests should be processed without an error. Capture the error returned, if any, for these requests. Get the percentage of error'd out requests.
3. During the span of the same hour, requests beyond 250K should return an error. Capture the error for a couple of these kind of request.
4. Gather similar response time statistics as in aforementioned ticket.
5. Run a few iterations of the test for 2 or 3 hours.
**To-Do**
[ ] create a `GitLab` repository that will hold the testing code
[ ] create a CI/CD pipeline that will run the code from the repository (pay attention to the runner tag)
[ ] trigger the pipeline to collect the test results
To consider
DV dependency