Location headers have wrong hostname when used behind load balancer

When you make a request like this that is routed through a load balancer

curl -i -u guest:guest-password -X PUT 'http://localhost:8080/gateway/default/webhdfs/v1/tmp/LICENSE?op=CREATE'

Knox currently will return something like this

https://backend:8443/gateway/default/webhdfs/data/v1/webhdfs/v1/tmp/LICENSE?_=AAAACAAAABAAAACQccBhGqTbDtfqAt7vzK1H39SnCZo7W14qCIs67ctZAJDXr9fEyJbo1H9AO8prLGdV8Jmz5TO_novslggJwY7E9Vep4eFP0auaxVpfBz4QG-ktSuviEU5aHl8om_SkuGLOwSDjBRZASXrV1huqKU-K_mKkCaPnC0NkCpRQRL0LMkGvB8yrl6_1vNkaoXTxwjm0kp1EhgniovHJVmfcPbjKmmoh-boVy1cj

To avoid confusion the 'backend' in the URL above is in no way correct but is in part caused because nginx is sending that value in the Host header. That is peculiar to nginx and could be fixed with nginx configuration.

The issue here is that Knox used the hostname from the Host header and the local port. I'm not exactly sure what the right answer it but I'm sure mixing is bad. We should either be using the information from the Host header or the information from the local endpoint of the socket. The way Knox was working before the fix for KNOX-439 was to use the local endpoint information so I'm going to fix this issue making that assumption.

I used nginx to reproduce the issue. This is the final configured I used to verify the fix. Note that the 'proxy_redirect' would need to be removed to see exactly what Knox is returning and compare to what is shown above.

worker_processes  1;

events {
    worker_connections  1024;
}

http {
    include       mime.types;
    default_type  application/octet-stream;

    sendfile        on;

    keepalive_timeout  65;

    upstream backend {
        server c6402.ambari.apache.org:8443;
    }

    server {
        listen       8080;
        server_name  localhost;

        location / {
            proxy_pass      https://backend;
            proxy_redirect  https://c6402.ambari.apache.org:8443/ http://$host:$server_port/;
        }

    }

}