Load or regressions testing with HURL

[matthiasschaub]

In the past, we tried to establish some kind of load testing using a python library called locust. This library is somewhat complicated and interactive, so it is seldom executed and instead manual checks are carried out.

OQAPI uses a tool called HURL to test for regression, which might also suit SMT better than locust.

What I would like HURL to test is a bunch of simultaneous requests to the marking detection endpoint (e.g. hurl --test --jobs 7 --repeat 7 stress.hurl). The SMT instance should be running in a Docker container with the same memory limits as in production. The SMT instance could be the test instance or a local instance. HURL should issue more requests than workers can handle simultaneously (currently more than 6) to test if memory is sufficient.