Hello,

1.14.0-1 running on Debian Stretch:

# dpkg -l | grep nginx
ii nginx 1.14.0-1~stretch amd64 high performance web server

I'm trying to load balance between two VMware View Connection servers (10.7.18.121 and 10.7.18.122) listening on 443/tcp, 4172/tcp and 4172/udp.
The way the application works is: first, the connecting client hits 443/tcp where authentication takes place, then the client gets connected to 4172/tcp (or 4172/udp).

I have no problems when the connection is handled by the same upstream server, such as: remote_client > nginx_vip > 10.7.18.121:443 > 10.7.18.121:4172 or remote_client > nginx_vip > 10.7.18.122:443 > 10.7.18.122:4172. However, I get application errors if 443/tcp is handled by one server and 4172/tcp/udp by another.

Therefore, I was wondering whether it'd be possible to configure Nginx in a such way that the upstream server is retained through the whole session? I mean, if a client gets served by 10.7.18.121:443 Nginx will use the same upstream to deliver 4172/tcp/udp?

I can probably switch to active-backup model, but I was hoping to benefit from the load distribution.

Many thanks.

stream {

log_format basic '$time_iso8601 $remote_addr '
'$protocol $status $bytes_sent $bytes_received '
'$session_time $upstream_addr '
'"$upstream_bytes_sent" "$upstream_bytes_received" "$upstream_connect_time"';
access_log /var/log/nginx/stream_access.log basic;

upstream test_horizon_4172_tcp {
hash $remote_addr consistent;
server 10.7.18.121:4172;
server 10.7.18.122:4172;
}

upstream test_horizon_4172_udp {
hash $remote_addr consistent;
server 10.7.18.121:4172;
server 10.7.18.122:4172;
}

upstream test_horizon_https {
hash $remote_addr consistent;
server 10.7.18.121:443;
server 10.7.18.122:443;
}

server {
listen 4172;
proxy_pass test_horizon_4172_tcp;
}

server {
listen 4172 udp;
proxy_pass test_horizon_4172_udp;
}

server {
listen 443;
proxy_pass test_horizon_https;
}

}