Here's a browser test illustrating the issue: https://crrev.com/c/1150146

This problem is not a prefetch specific, right?
Also, since the g_max_sockets_per_group is a global limit, even if we have a per-tab simultaneous same-origin prefetch limit, it still can consume all available sockets among multiple-tabs.

So, adding such a limit sounds an incomplete and ad-hoc approach to me.

> This problem is not a prefetch specific, right?

It's not. A page might just manually send a bunch of XHR requests to an origin causing the same problem. But we could add special treatment for prefetch requests though.

> Also, since the g_max_sockets_per_group is a global limit, even if we have a per-tab simultaneous same-origin prefetch limit, it still can consume all available sockets among multiple-tabs.
> 
> So, adding such a limit sounds an incomplete and ad-hoc approach to me.

We could enforce this limit among all tabs, for example at the ResourceScheduler level.

Alternatively, we could just stop all prefetch hints after the user initiates any kind of page load. Mozilla claims they do that: https://developer.mozilla.org/en-US/docs/Web/HTTP/Link_prefetching_FAQ#What_happens_if_I_click_on_a_link_while_something_is_being_prefetched.3F 

Prefetch is a speculative fetch (not mandatory) so technically we can surely implement this I suppose (in global way too, the latest architecture allows this).  Marking this Available for now, I'd like to get more inputs/signals/feedback before we actually implement this.