Request time doesn't include download time.

[bradleyayers]

From my testing it seems like the time reported by locust doesn't include the download time of the response. This assumes that the server only begins sending the response after the entire response is buffered.

I couldn't find anything in the documentation to clarify what the expected behaviour is.

[fordhurley]

Is it possible that you're passing stream=True to the client.get call? If so, then you definitely aren't including downloading the content in timing. By default, this is False, so that time should be included in the reported time. stream is passed through to requests, which uses it like so.

Got this by looking at the locust and requests code, so I could be wrong.

[Jahaja]

Closing this as no further response was given. Furthermore, Locust uses non-streaming requests by default.

[heyman]

I added a test in f05ff7f which shows that this is a real issue. I didn't realize it at first, since I was running requests 2.2, but saw that the build failed on Travis.ci (https://travis-ci.org/locustio/locust/builds/34592126). The test starts failing with requests 2.3. I haven't had time to investigate what might have changed yet.

If anyone is affected by this right now, a work-around is to downgrade to requests 2.2.

There's also a few other tests that fail because of ProtocolError exception being raised. This is however a bug in requests 2.4 which has now been fix: https://github.com/kennethreitz/requests/pull/2193

[Jahaja]

I guess you mean https://travis-ci.org/locustio/locust/builds/34596048? The streaming test is currently failing on AttributeError: 'TestHttpSession' object has no attribute 'assertGreater'.

Edit: Ah, I just looked at the first one which is 2.6 that does not come with assertGreater.

[heyman]

Ah, that's the python 2.6 test suite that is failing because of that. The 2.7 one (https://travis-ci.org/locustio/locust/jobs/34596050) is "a real" failure because the content download time is not included in the response time (with requests 2.3 and 2.4).

I've fixed the issue under python 2.6 in e18bf5c.

[Jahaja]

Yeah, looks like it. Nice find.
We could use unittest2 instead of copying the methods as we already includes it in the tests_require option in setup.py.

[heyman]

@Jahaja: Ah, okay. Yeah, that sounds better :).

Since this issue is caused by changed behaviour in requests, I've created an issue (with a similar test included) there (https://github.com/kennethreitz/requests/pull/2209). I believe the changed behaviour in requests was unintended, but if it turns out it wasn't I guess we might want to do some work-around in Locust.