Handle backlog on UNIX socket too

[pinaraf]

The cheaper subsystem specifies that "backlog is only available on Linux and only on TCP sockets (not UNIX domain sockets)." This commit specifically implement this: UNIX domain socket support for backlog on Linux, using Netlink to call the kernel and get the queue status.

[pinaraf]

Note: this has been successfully tested locally, on a debian testing system, with the following:

WSGI app:

#hello.py
import time

def application(env, start_response):
    time.sleep(10)
    start_response('200 OK', [('Content-Type','text/html')])
    return [b"Hello World\n"]

uwsgi launch:

uwsgi --http-socket /tmp/uwsgi.sock --plugin python3 --module hello --master --processes 500 --plugin cheaper_busyness --cheaper-algo busyness --cheaper 5 --cheaper-initial 10 --cheaper-overload 20 --cheaper-step 2 --cheaper-busyness-multiplier 10 --cheaper-busyness-min 20 --cheaper-busyness-max 70 --cheaper-busyness-backlog-alert 16 --cheaper-busyness-backlog-step 2

curl to fill the uwsgi queue:

for i in `seq -w 1 100` ; do curl --unix-socket /tmp/uwsgi.sock http://127.0.0.1:9090/ &> /tmp/curl$i.txt & done

extract from uwsgi output with the patch:

[busyness] 72 requests in listen queue, spawning 2 emergency worker(s) (20)!

Without this patch, the previous output never shows, showing the backlog alert system is broken, and thus the recommended settings are currently unsafe: for cheaper-overload, a high value is recommended, but it can not be properly used if there is no backlog alert to catch load spikes.