Connections to the fast cgi server

Dimensions in which we can move:

• number of concurrent tcp connections to the server
• number of concurrent requests per tcp connection (requests/responses can be interleaved like H2)

Sharing a vconn/tcp connection between different requests will cause us to have a lot more synchronization and this is not implemented yet.

What is best practice right now in nginx for connection management? Focus on # connections or # requests per connection or mix?

A way to get a first implementation working is to have a limit on the number of concurrent tcp connections and not implement the interleaving yet...

This will mean we will have to have queueing in place for those requests...

How and where (at which callback) are we going to queue a request for the backend (before TCP connect obviously), and are we going to disable the incoming vconn?