Nginx nginx 1.13.9 - released

Dont recall the exact reason but IIRC it might have been a GCC compiler thing I see you used Clang

edit: ok it was Clang related Beta Branch - set -Wno-error=strict-aliasing when NGINX_HPACK='y' set IIRC was testing with Clang 3.4.2 so could be related you're using Clang 5.

 

What do you mean exactly conflicts ? You mean there's no HTTP/2 Push preload hints headers shown if using fastcgi_cache ? Or fastcgi_cache doesn't work if HTTP/2 Push reload hints headers are used ? Did you setup fastcgi_cache to ignore and bypass cache-control and cookies ? Or only specific cookies to ignore ? You might want to add the HTTP/2 Push mapped cookie to fastcgi_cache ignore to have HTTP/2 Push + fastcgi_cache together but don't see the point as HTTP/2 Push would be best for first time visitors only and not repeat visitors as they would already have the pushed asset in browser level cache on repeat visits.

If you setup cookie mapping for HTTP/2 Push preload hints headers, then HTTP/2 Push will only ever work on first time visitors and repeat visitors get the cookie and skip HTTP/2 Push in favour of whatever default caching you using - be it browser level cache or something like php's fastcgi_cache. Well that's the theory as I haven't tested with fastcgi_cache. Best to use webpagetest to test HTTP/2 Push vs fastcgi_cache and see for page load speed WebPerf - PageSpeed - How to use webpagetest.org for page load speed testing. Either way ideal is for repeat visits using fastcgi_cache without HTTP/2 Push anyway as that in theory is fastest for repeat visits.

Yeah renamed it so not confuse zlib 1.2.11 custom library with zlib 1.2.3/1.2.8 system zlib library

 

Use Chrome 64 to test not Canary seems initiator column doesn't show PUSH sometimes but if you check Canary chrome://net-internals HTTP/2 section it registers the push. So Chrome 64 more reliable if you're checking via devtool network's initiator column.

Also cookies can be cleared directly in devtools via Application > Cookies section

Well that's the ideal way, repeat visits should ideally not use HTTP/2 Push as it will be slower than from browser level cache or any other caching you setup for repeat visitors. The distance traveled between visitors browser cache to browser is much shorter than from your server cache to visitor's browser. HTTP/2 Push ideally should be used on first time visitors only.

On my test HTTP/2 demo site in California to me in Brisbane Australia, the HTTP/2 pushed CSS downloads in ~13-16ms vs non-pushed first time visit CSS downlaod at 198ms vs browser cached repeat visit CSS download at 8-9ms.

So ideally you want first->repeat being 16ms->9ms rather than 198ms->9ms

 

Oh i see so not the same visitor but all visitors who visit and get fastcgi_cache'd page. Might want to report the issue to Nginx Trac Tickets for 1.13 – nginx

Oh you might want to check your fastcgi_cached page if it cached your push mapped cookie too and/or maybe adjust your fastcgi_cache_key Module ngx_http_fastcgi_module

 

Cheers Owen you're welcome. I'll get around to testing web fonts and HTTP/2 push too. Interesting to see how different web browsers handle it

 

You're using comodo SSL cert on which domain ? on http2.domain.com and self-signed ssl cert on upload.domain.com right as intended right ? How was upload.domain.com Nginx vhost site created ? via centmin.sh menu option 2 with self-sign ssl option = yes ? If you setup comodo manually on http2.domain.com vhost, you might have misconfigured it. Might want to start a new thread in Domains, DNS, Email & SSL Certificates for this.

 

Yeah cache aware push should solve the wastefulness though I doubt any live site doesn't run with 3rd party assets - Google Adsense and Google Analytics being 2 of the most common. I just wish Google would look at speeding up Adsense and Analytics !

 

@Sunka start a new thread in Domains, DNS, Email & SSL Certificates for this

But if you only have 1 domain with ssl why are you curl header checking http2.domain.com ? You want to be curl header checking a valid HTTPS site with valid SSL certificate

For example I specifically created a new Nginx vhost site http2.centminmod.com via addons/acmetool.sh with default Letsencrypt SSL certificate for testing Nginx HTTP/2 demo

Code (Text):

curl -I https://http2.centminmod.com/
HTTP/2 200
date: Wed, 21 Feb 2018 13:42:33 GMT
content-type: text/html; charset=utf-8
content-length: 6103
last-modified: Sun, 11 Feb 2018 01:14:19 GMT
vary: Accept-Encoding
etag: "5a7f98eb-17d7"
server: nginx centminmod
x-powered-by: centminmod
strict-transport-security: max-age=31536000; includeSubdomains;
referrer-policy: strict-origin-when-cross-origin
x-frame-options: SAMEORIGIN
x-xss-protection: 1; mode=block
x-content-type-options: nosniff
set-cookie: pushcheck=pushed
link: </purecss/pure-min.css>; rel=preload; as=style, </purecss/grids-responsive-min.css>; rel=preload; as=style
accept-ranges: bytes

 

Yeah usually less number of files + smaller sized files will win regardless of how it's being served. I remember nginx + ngx_pagespeed being faster in both HTTPS and non-HTTPS page load tests than nginx + HTTPS (SPDY/HTTP/2) as ngx_pagespeed bundled up the css and js. That's why I like ngx_pagespeed too

Though there's a slight difference between total page load times versus perceived visual render times (TTFB, Time to first meaningful paint etc). Will be doing alot of Webpagetest.org tests for this WebPerf - PageSpeed - How to use webpagetest.org for page load speed testing

Oh and remember to test on slower speed net connections like 3G mobile profiles

 

Yeah if you haven't checked out the HTTP/2 Revisited Youtube by Hooman Beheshti from Fastly at New York Web Performance Meetup then check it out. It also talks about HTTP/2 Push adoption and it's very very very and I mean very low right now SSL - WebPerf - PageSpeed - HTTP/2 HTTPS Revisited. LiteSpeed/OpenLitespeed, Caddy and H2O are all HTTP/2 web servers with HTTP/2 Push but their market share is low compared to Nginx so might account for adoption rates right now.

Now that Nginx has HTTP/2 Server Push, might see that adoption rate rise