On large amount of requests keter may run out of sockets

[jappeace]

We were being hit with hundreds of requests a second, with some of them taking longer than others - the result being that eventually the system ran out of sockets (aka file descriptors) to service all the requests, and Keter fell over.

So keter should be able to deal with this.
But once this happened it also didn't reboot (which is a nixos config issue, which is now also part of this repository).

[jappeace]

This is reproducible on massive resource usage from the app behind keter.
But we worked around this by using a ratelimit in nginx, eg:

     limit_req_zone $binary_remote_addr zone=applimit:10m rate=10r/s;

and then in the vhost:

             limit_req zone=applimit burst=20 nodelay;

For our app it becomes next to impossible now to get to the run out of socket
scenario, because we also use allocation limits for every request