Very low Start Render times - error?

Recently, we’re seeing some very low numbers on Start render time, which doesnt seem realistic, wondering if its a problem in WPT, or if Chrome has changed the way it reports Start render time?
Start Render occurs before download of html document is complete, before css files have even started download, around 200ms. Normally, our Start Render is around the 2s mark.

Example: WebPageTest - Visual Comparison

We are using akamai edge start to get the very fast ttfb, which only includes HEAD element with css refs and dns prefetch, then the rest of the html follows later, with download complete around 500ms.

How can Start Render occur before html body is downloaded?
Has anyone else seen the same thing?

[attachment=558]

Can you reproduce it on the public WebPageTest instance? Do you see it happening all the time or intermittently? Looking at the filmstrip looks like Chrome is drawing something along the bottom of the screen and then flashing black to a blank screen which is triggering the false start render time: https://wpt.speedcurve.com/results/16/01/30/QC/70fce96f18d53d96599147a8092cf4f7/video_3/frame_0002.jpg

If I can get it to reproduce I should be able to fix it easy enough but I’ve never seen it happen before.

I’m seeing the same issue with different pages, it seems Chrome is adding the grey “waiting for [page]” message at the bottom of the page and WPT is considering this as the start rendering of the page on the screen which is false.

Do you have a link to a sample test? There is a margin on the edge of the page that gets ignored when looking for changes in screen content so it shouldn’t be tripping it up but it’s possible that the size of the message changed or something else is going on. That’s different from the issue above which is displaying some other content.

I have a sample showing the same black background at the bottom of the page and have other samples just showing the “waiting for…” Chrome message, I’ll send them to you directly due to confidentiality reasons.

I think I managed to reproduce it on the public EC2 Frankfurt instance and just pushed a work-around that should address the issue (couldn’t reproduce it with the fix but it was intermittent before).

Hi , I believe one of my colleagues may reach out independently/directly, but for everyone’s benefit, I would like to ask on the forum since I’m a noob on the admin/install side of WPT:

  • We’re running private instances as well as our clients. How can we get the patch, is it publicly available/checked in ?

Thanks for your help on this, much appreciated!

Here is the info on getting the latest agent: https://sites.google.com/a/webpagetest.org/docs/private-instances#TOC-Updating-Test-Agents

If you are running a recent web UI you can also add a setting to settings.ini to have the server automatically update the agents as I release them: webpagetest/settings.ini.sample at master · WPO-Foundation/webpagetest · GitHub

Hi Pat, I’m not sure if this is related to the start render issue Allan flagged on http://www.webpagetest.org/forums/showthread.php?tid=14199
but I’m seeing a graphic or thin line appearing in webpagetest at 0.2s/0.3 regularly on a chrome agent. It’s badly skewing start render metrics. for mobile it almost looks like a phone outline

It’s happening on several sites, by different customers, so I don’t think it’s the sites, but maybe the agent. It also seems to be an EU agent all the time, but that might be a red herring (or not)

Allan’s issue seems to be a loading message, but this doesn’t seem to be the case.

If you’re willing to look at this, I can send you some screen grabs. Unfortunately it’s a private/firewalled instance so I can’t send you a link.

The artifact is different between desktop and mobile and does not look like a loading message

We are seeing this impossibly low start render time quite a bit as well, Pat:

http://www.webpagetest.org/result/160218_A3_Q6M/7/details/

http://www.webpagetest.org/result/160218_ZT_6BH/4/details/

It’s persistent at the Sydney node.

The filmstrip shows a RESTART PC message… it happens right away, so I guess whoever maintains that machine needs to patch

Ah!

Seeing it elsewhere, though. I’ll try to dig up some tests…