On 2021/02/16 20:12, Franz Bettag wrote: > My point, the protocol after HTTP 1.0 encourages keep-alives anyway. > Close is only default in 1.0 so basically you wouldn’t have lingering > dead sockets on your server.
If you want a full featured HTTP implementation then perhaps relayd is not for you. It is an amalgamation of a web firewall type of thing with a load balancer / dead backend detection so any diff needs to consider all use cases. It is not ready to handle multiple requests in one connection, it makes a decision about where to send the request then gets out the way (handing the packet-shovelling over to the kernel via "splice"). To handle persistent connections it would need to stay in the loop for future requests and process them separately, maybe to a different backend. Anyway I am not all that interested in relayd http proxying so I will drop out of this thread, just trying to give a clue about what might be needed in order to move this ahead. I am not saying the diff *is* problematic but you can bet anyone committing it will need to satisfy themselves that it doesn't introduce a new problem and there has not been discussion of that in the thread afaicr.