Fix problem with docker `internal` networks

[anlorn]

Hello, I found an issue while setting up nginx-proxy container for my network.

I have the following setup:

1. network 'A' is a regular network created in docker
2. network 'B' is an internal network created with --internal parameter

Container with nginx-proxy connected to both A and B networks. At the same time, containers with applications are connected only to 'B' network. When running such a config, I was always getting.

After investigating, I find out that the issue comes from the fact that 'docker inspect' section NetworkSettings.Ports is always empty for internal networks, no matter which port you expose/publish. As the result we always have empty addresses variable when we generate a template. Solution here is to take exposed ports from section Config.ExposedPorts when NetworkSettings.Ports is empty.

How to reproduce: Create internal network docker network create 'test_network' --internal Run httpd container connected on this network docker run --network test_network --rm -e VIRTUAL_HOST=test.local -e VIRTUAL_PORT=80 httpd:2.4-alpine Get latest nginx.tml wget https://raw.githubusercontent.com/nginx-proxy/nginx-proxy/main/nginx.tmpl Run docker run --network test_network -it --rm -v /run/docker.sock:/tmp/docker.sock -v $(pwd):/app/ jwilder/docker-gen /app/nginx.tmpl You will see that in the upstream section of the generated config server directive is marked as down. Repeat all the steps but this time, on step 1, create a network without the --internal parameter.