Hi @Jdlrobson thank you for creating the issue, I've missed it. It looks like there could be some problem with the WebPageTest instance. I could see that the issue is for all URLs we test for emulated mobile and there's an increase in TTFB. I wonder if something changed on that instance, I'll look into it.
I've checked where we use WebPageReplay to replay the traffic (=no variation in TTFB, only measuring frontend) and there it all looks good:
Also checked my own instance and there's nothing there either than the usual noise:
I'll check our WPT agent and see if I can see what's wrong.
FYI the last visual change is the banner, you can see that if you check the screenshots of the runs. It will change SpeedIndex and LastVisualChange.
Start render should never change though, so there something is broken.
Comparing waterfalls it's a big gap between before and after:
It reminds me of a bug we had running Firefox on a too slow instance years ago, but then it should affect other metrics too.
Comparing a run today and one from September 2, I see the exact same requests before reaching the first render. The bandwidth graph up to that point is pretty much the same as well. The only different seems to be that the browser main thread just takes a lot longer to run before reaching the initial render:
Digging into the processing timeline, it seems like it's the browser's first layout taking 1.1 extra seconds:
Comparing the HTML of the "last good run" (04/09/2019 12:01:17) and the "first bad run" (04/09/2019 14:11:02), it's absolutely identical.
Looking at the details of the other requests, the only difference I see is that 220px-President_Barack_Obama.jpg becomes a webp in the "first bad run".
Doing the same comparison on the ruwiki Facebook article, all the responses are perfectly identical except for the HTML. In that case, the difference is the MediaWiki version (first bad run with 1.34.0-wmf.20) and some minor DOM difference where attribute values appear:
And this body class is removed:
And there are related config changes in the head JS.
AMC going away:
And menu-related stuff being added:
For the Sweden article, however, the HTML is identical in the last good run and the first bad one. Making the HTML change above a red herring.
Once again, when looking at the processing timeline, the bulk of the extra time is spent on the initial layout.
I think we can conclude that the issue is in the testing environment, where the browser is somehow much slower than before for emulated mobile's initial render.
The device emulation is a bit of a black box. It might be good to verify that there wasn't a system update on the host around that time or something like that. I imagine that updating some linux library on the system could have a side-effect like this for example. Or a hardware incident of some kind on the host.
I've tested on a fresh agent (same size) and it looks the same: http://wpt.wmftest.org/results.php?test=190910_SB_9b8bf08859c512614318c497b335b0b2
I also saw this today: https://github.com/WPO-Foundation/wptagent/issues/285 - problem for users running emulated mobile on WebPageTest, looks upstream Chrome changes. Wonder if a Chrome update has been pushed that caused it? On the other instances we run the exact Chrome version all the time (+ we don't use the netlog to create the waterfall).