FaaS is closer in mindset to CGI than fCGI : if you're gonna have a server accepting connections, you might as well accept HTTP connections. CGI and FaaS are more like developing a function that transforms a request into a response, and let the infra deal with the rest.

It is true though that it would be nice to have a standard for FaaS that doesn't have the inefficiency of CGI.

Nah, it's FCGI because the process is kept "warm" and reused. It's much like the typical execution model of php-fpm.

That the process is kept warm is an artifact of the infrastructure. As a Dev you absolutely don't care about how it's run and that's the whole point: just implement your business logic and the system takes care of the rest. As a dev it's all transparent.

That's actually less true of Lambda than of php-fpm. In Lambda, you can mutate the state of global variables between requests[1]; you have to take care not to do that. In php-fpm, all is thrown away.

[1] http://blog.matthewdfuller.com/2015/12/aws-lambda-occasional...

That's true for php-fpm, though many use something like apcu to cache state between requests. FCGI itself doesn't mandate that "thrown away globals" behavior. Golang under FGCI, for example, just spawns a goroutine for each new request.