Waterfall Weirdness

Okay… There is probably something really simple that I’m overlooking, but I’ve pulled out enough hair trying to find it so I figured I’d ask while I still have eyebrows.

I’m running a stand-alone private instance server and three agent machines - 1 IE-9, 1 IE-10 and one IE-11/Firefox/Chrome. All are Win7 Pro, 2GB VMs running WPT 2.15 under VMWare 4. I have a script that exercises several pages on the site and I just turn ‘logData’ on and off for the page I’m interested in. Using both a local copy of PageTest and the server-based URLBlast for IE-9, everything works as expected. With IE10 and WPTDriver, the script runs thru to the end (after a few tweaks) but hangs at the exit page (presented after the Log-out button is pressed) and eventually times out (no repeat view stats). IE11 we won’t even talk about. It makes it two-thirds of the way through one time, only about five actions the next and all the way to the log-out hang (like IE10) the next. FWIW - the log-out step is very simple:

   clickandWait   className=logout

That operation works. It actually logs out of the site, but it then just sits on the exit page (when using IE10 or IE11) instead of closing it and moving on to repeat view. I’ve tried just a plain ‘click’, I’ve tried setting a DOMElement check and even plain 'ol sleep delays but nothing seems to want to pass beyond this point when running with WPTDriver.

Since my main curiosity was the log-out hang, I modified the script to simply log on and log off; capturing data only for the exit page. And that’s where the weirdness came into play. Notice in the waterfall view below, for IE-9 under URLBLast there is a nice, typical page progression:

Render Start > FirstPaint > DOM Load > On Load > Document Complete.

With IE-10 Under WPTDriver, the progress is:

Document Complete > FirstPaint > DOM Load > On Load > Render Start

I’m not sure, but that just doesn’t look quite right. So I gave it a shot with IE-11. Under IE-11 and WPTDriver (not shown) it’s an even simpler chart - a Document Complete at around 0.032 and a Render Start at 0.16. Nothing else. Don’t know if it means anything, but I did notice that the relative position and timing of the elements is pretty much the same on all testers. Well, just for the fun of it, I decided to see what IE-9 under WPTDriver would look like (bottom image). Sorta similar to IE-10, with a pattern of:

Document Complete > FirstPaint > Render Start > DOM Load > On Load


I’ve read some of the discussions re: fast load times overrunning the browser’s display capability, but I would have thought the IE9 runs should be reasonably similar. Particularly since the IE-9 URLBlast and IE-9 WPTDriver tests were both run on the same machine.

Any suggestions on this one? (I’ll save my IE10-11 script issues for another post).

Thank you


Don’t worry about the render relative to the onload, that won’t cause the hang/timeout. It’s quite possible that the actual click for the logout isn’t working as you’d hope/expect in wptdriver. What would probably work best would be to try doing it in Javascript from a dev tools console and once you have the JS necessary to trigger the logout use an execAndWait instead.