Support for custom time metrics

[mdomans]

Hi, recently I needed to load test an elasticsearch cluster and for what it's worth, I wasn't interested in connection time, only in the work time spent by the ES server. Elasticsearch returns that data as part of a JSON response, under a took key. I wanted to measure that, so I hacked this:

if response.status_code not in [200, 300, 301, 302, 303]:
    response.failure("Status different from 200")
else:
    resp = json.loads(response.content)
    events.request_success.fire(
        request_type=response.locust_request_meta["method"],
        name=response.locust_request_meta["name"],
        response_time=resp['took'],
        response_length=response.locust_request_meta["content_size"],
    )
    response._is_reported = True

I think it could be wrapped in something much nicer, e.g.:

if response.status_code not in [200, 300, 301, 302, 303]:
    reporting.record_failure(response)
else:
    reporting.record_success(response, took=10)

I was interested if anyone else is interested in being added to locust

[heyman]

Hi!

Rather than modifying Locust, I would probably write a small custom client (which wraps python-requests or some other HTTP client) that fires the request_success & request_failure events.

I'm not sure I understand what you're proposing. Is it the API for fire:ing request_success/request_failure that you think could be simplified?

[mdomans]

I think I understand your approach better now. I think you lack good docs on how one should approach doing such custom stuff, though I've got the idea now how to port my code. Can I contribute it as an example?

[justiniso]

@mdomans yes, definitely!