Fix NGINX regression Docker Expose

[Alkarex]

#2721 introduced a regression by not properly declaring the exposed port in Dockerfile.
The Apache version did it correctly:
https://github.com/docker-library/php/blob/faf8864e3845ced80780c03eefc66c022e2f9ac1/8.1/buster/apache/Dockerfile#L289

The consequence is that other systems, e.g. Traefik, could not know what port to route to.

[dvikan]

Why can't Traefik forward requests to 127.0.0.1:80? I don't understand the purpose of EXPOSE. As far as I can tell its purpose is internal container communication?

Also keep in mind that we have a configuration setting for changing the port that nginx binds to.

In summary, maybe our docker setup isn't properly written for real production use.

[Alkarex]

Of course it can be done manually and/or overridden but with a default EXPOSE, routing can be done automatically without having to hunt in the documentation which port each service is using internally.

You can for instance try a docker ps with and without the EXPOSE instruction and observe the difference.

Not related but just for the record, 127.0.0.1 would not work because localhost in a container is not the same localhost in another container.

[Alkarex]

Reference for Traefik, but it is not specific to it:

https://doc.traefik.io/traefik/providers/docker/#port-detection

[yamanq]

Good catch, thanks for contributing! I won't merge this yet in case @dvikan has some comments.

[Bockiii]

You are correct. I'm running this in docker-compose with defined ports, but in a helm scenario, the automated port assigning wouldnt work.

I'm approving this change, but there will probably be another change in the near future because I want to switch to the lsio nginx container. So thanks!