Samuel Williams

> First we have a pretty variable workload and though we can predict how many proxy servers we are going to run by way of capacity planning, having what is...

> P.S. I do believe that we could achieve this throughput if it were possible to get access to the nio4r socket objects from within falcon already tho. Realistically, the...

Can you explain your ideal slow client disconnecting policy? e.g. less than x bytes/s for y minutes? or something else?

This is something I’ve thought about for a while. If you call `socket.read` should that operation block indefinitely? I think the answer is no. Especially not by default. There should...

Unfortunately it's not sufficient. It needs to be in the buffering/socket layer.

You need to handle the timeout. Something like ```ruby begin task.timeout do socket.write(...) end rescue body.close end ```

If that's the route you want to go down, even just temporarily, you should probably make a body wrapper with this behaviour. But as I said it's not a fully...

Without digging into it too much (dude I'm on holiday at the beach), I did a quick test of your code vs using then `LimitedQueue` and got a 3x perf...

I think a timeout at the socket level makes sense for almost all protocols. I don't even know if the timeout allows to be reset but something like this would...

@julik I'm playing around with some timeout/throughput concerns. - Do you think it makes sense to have per-socket timeout for any/all async operations? e.g. connect, send, recv, read, write, etc....