incomplete waterfall with firefox

[nilskuhn]

Hi Pat,

we observe strange behaviour of a simple test when measured with firefox (no matter which version).
The testscript:

Code:

logdata    0
navigate    https://beta.otto.de
logdata    1
navigate    https://beta.otto.de/damenmode/

In IE8+9 and Chrome results are like expected:
http://www.webpagetest.org/result/130529...e1feb57f3/
http://www.webpagetest.org/result/130529_S8_XPJ/

Same script, run with FF delivers reproducible incomplete waterfalls. Doc ready-time (which we are interest in most) could be correct. But there is a negative (?!) start render-time and we don't really trust in the measured doc ready-times...
Examples:
http://www.webpagetest.org/result/130529_RK_XBW/
http://www.webpagetest.org/result/130529_Y5_XFA/

we also tried the following:

• replace navigate with click(AndWait) or exec(AndWait) with javascript
• sleep 15 in front of logData 1
• additional navigate http://google.de in front of logData 1
  

Without any success.

Without logData 0/1 and combineSteps we have full waterfall of both navigates:
http://www.webpagetest.org/result/130529...1/details/

Any ideas?

Regards Nils

HINT: Please contact me to get the whole script of failing tests cause there are some cookies that should be set.

[nilskuhn]

no idea?

[pmeenan]

Sorry, I was in the middle of reverting GPU support on the VM's which was on for about a week and wanted to see if maybe that was causing the issue. Do the tests look better now? http://www.webpagetest.org/result/130605_32_1H8T/

What are you expecting to see in the 2nd navigation?

[nilskuhn]

I expect to see the second part of the combined-step-measurement: So approximately requests 98 to 142 of http://www.webpagetest.org/result/130529.../details/. Or something similar to an IE-result of the same script: http://www.webpagetest.org/result/130606.../details/.

Your new result is better in that it starts where we set logData 1, so with the second html-request. But the waterfall is still incomplete and the doc ready time of 0.873 secs isn't correct, too (would be great).

[nilskuhn]

I've tested with different versions of the agent-(wptdriver-)components:
* agent-/wptdriver-version 77 (WPT 2.8)
* agent-/wptdriver-version 87 (WPT 2.9)
* agent-/wptdriver-version 96
* actual agent-/wptdriver-version 116

For the older versions 77 and 87 I had to downgrade FF to version 17 because with version 21 I just got blank page and tests didn't start.
I've incomplete waterfalls in all versions with the given script. The result-fie 1_IEWTR.txt on server already contains too less requets, so the measurement itself seems to start too late (after setting logData 1). It's definitely no problem with the newest version.