Reverse Proxy with 500k connections

[larsg]

Hi everybody,

ok, I recognized another linux network problem that I solved now. Situation
now is like following:

When I call my upstream address via curl (on the nginx host) by selecting
the corresponding local interface (eth0-9 = 192.168.1.130-139) everything is
fine.
    curl https://192.168.1.21:443/remote/events --insecure --interface eth0

But when I specify the same IP address that refers to eth0 (eth1-9 etc.) I
get an "110: Connection timed out".
Does anybody know this situation? Checked by sysctl config but it looks
fine...

split_clients "${remote_addr}${remote_port}AAAA" $proxy_ip {
    100% 192.168.1.130;
}

server {
  listen              44301 ssl backlog=163840;
  proxy_bind $proxy_ip;
  #proxy_bind 192.168.1.130;
..

2017/03/09 16:54:33 [error] 30081#0: *11 upstream timed out (110: Connection
timed out) while connecting to upstream, client: x.x.x.x, server: , request:
"GET /remote/events HTTP/1.1", upstream:
"https://192.168.1.21:443/remote/events", host: "xxxxx:44301"

Posted at Nginx Forum: https://forum.nginx.org/read.php?2,272808,272858#msg-272858