Hello there, I have been seeing some odd behavior where “Fully Loaded” time is often long after the Document Complete time, even for sites with no javascript.
My test: http://www.berkshirehathaway.com/ (since it only has 3 objects, it eliminates a lot of external factors)
Results: http://www.webpagetest.org/result/120501_6C_4676S/
In Run 1 ( http://www.webpagetest.org/result/120501_6C_4676S/1/details/ ), Document Complete and Fully Loaded are fairly close, and the difference appears to be the nonexistent favicon.
In Run 2, however ( http://www.webpagetest.org/result/120501_6C_4676S/2/details/ ) we get a Fully Loaded time of 2.5 seconds.
Run 3 is similar to Run 1.
Why is the result in Run 2 so far off from the other two? It throws the median quite a bit, and I can’t figure out what would still be loading in Run 2 that would make it still see network activity.
I have seen this in quite a few different tests, but this is by far the simplest example. Am I just missing something about the way Fully Loaded is calculated?
Here is another example: http://www.webpagetest.org/result/120502_PF_46NT7/
In this one, it looks like all the tests are showing a fully loaded time well after the waterfall stops. This is a much more complex site than the other test, and it appears that the Fully Loaded results are more frequently in the 4-5 second range even though the waterfall usually stops at 2.5-3.2
I frequently, but not always, get this type of result with this particular site. That’s what led me to check other sites… I thought there was something special about this site that was still loading behind the scenes (though from looking at network traffic while loading the site, I haven’t yet seen anything).
The Berkshire Hathaway site, however, has only three objects, no javascript, no SSL, and it’ll occasionally do it too.
Ahh, thank you for a reliable failcase. I captured a timeline for the site and there is nothing in their either: http://www.webpagetest.org/result/120503_SR_3986a0bf2e8d4cd68c464f8593b55fb0/
Now that I have a site that reliably shows the problem I should be able to track it down pretty quickly (absolutely looks like a problem with the Chrome agent).
I opened a critical bug to track it: Google Code Archive - Long-term storage for Google Code Project Hosting.
I’m expecting to be in the Chrome agent code today adding support for seeing the SSL timings so I’ll get it fixed while I’m in there.
Hi Pat – I was looking into this issue again, and I couldn’t tell from the SVN commit notes or issue tracker if it’s been changed.
For my private instance, should I go with the development copy of wptdriver, or is there an upcoming 2.7 private instance I should wait for?
Just fixed it. I have one other fix I want to get in and then I’ll release a new binary this afternoon.
ok, you can grab the updated agent from here: http://www.webpagetest.org/work/update/wptupdate.zip
To install the update:
- Copy the zip file to your /work/update folder on your server
- Extract wptupdate.ini from the zip file to the same directory
The agent will update itself before processing any more tasks.
Thanks,
-Pat