103 early hints results

[adardessign]

We just implemented 103 early hints on our product detail pages
the results look pretty interesting

There are 2 issues

1. there are occasional (-1) in the waterfall what is that spouse to mean?
2. the performance entry TTFB for the main page is now broken as it measures the early hint resource

Here is a link to one of the tests

WebPageTest

[pmeenan]

-1 usually means that a request was canceled or didn’t complete but that doesn’t look like the case here though maybe it is because 6 and 7 look to be the same or similar resources (though so do 2 and 3 so maybe something funky is going on with the request processing.

For better or worse, the early hints impacting TTFB is expected and to spec for navigation timing. It could be useful to have a separate TTFB that measures the time to the first byte of payload data.

[adardessign]

Thanks!
The similar looking (-/) resources (requests 2-3) are CSS but requests 8-12 are JS

Trying to understand why if there are no blocking resources it takes an extra 500ms to get to LCP

close up waterfall 1

close up waterfall 2

I’ll create a separate thread for TTFB that measures the time to the first byte of payload data when using 103 early hints

Thanks again

[pmeenan]

If you look at the dev tools timeline view it can probably tell you what script code is executing there but from the look of the pink chunks in the waterfall it is probably the same inline code that runs at ~1.4s in the second waterfall and timing just happens to have it running before LCP.