> So in the case of the toot mentioned above there is an AAAA for the FQDN. But the connection doesn’t work. But the end user doesn’t notice. Because there is a browser feature called Happy Eyeballs (HE). Basically: A browser tries both protocols and uses the faster on. In case of non-working IPv6 IPv4 is always faster.

I mean even if there is no "Happy Eyeballs", it would still work for end users, no? Unlike `curl`, you don't specify -4 or -6 with browsers. Without HE, wouldn't (shouldn't?) it just try both in sequence?

> HE can have some funny side effects. In a project a connection to a development web server sometimes worked and sometimes didn’t. The solution was quite simple. The customer used a split VPN tunnel. IPv4 was routed via the VPN tunnel and those IPv4 addresses were allowed in the web servers access list. IPv6 was routed via the normal Internet connection and those addresses weren’t allowed.

This following sentence makes no sense, when the author literally said HE help it work when one IP family was broken and the other is not. So why would it work sometimes if IPv6 is consistently broken and IPv4 is not? (I'm not saying it won't happen; just that it shouldn't be "caused by HE" -- HE literally is the solution to this issue in general as the author said themselves!)