Try a blocking write before doing async write call

[gbaraldi]

On a simple http benchmark this is a 7x throughput improvement.

[gbaraldi]

Motivating thing was https://github.com/AbhimanyuAryan/c_http_jl/ the remaining differences are, we always do async reads which might not be optimal, but libuv doesn't have a try_read (even thought it might eventually). The other is node seems to go into the writev call while we fall into write, and we plainly do a lot more write calls than they do. Not sure how much of this is on HTTP.jl and how much is on base.
Also HTTP.jl calls nagle lots of times while node doesn't seemt o call it very much at all

[giordano]

On a simple http benchmark this is a 7x throughput improvement.

Can we actually see the benchmark?

[gbaraldi]

const ROUTER = HTTP.Router()

HTTP.register!(ROUTER, "GET", "/", req->HTTP.Response(200, "Hi"))
# You can register more
HTTP.register!(ROUTER, "GET", "/bye", req->HTTP.Response(200, "Bye!"))
HTTP.serve(ROUTER)

and using https://www.npmjs.com/package/autocannon
It went from 100k requests to 700k requests on my laptop

[IanButterworth]

Build doesn't seem happy