Use server 127.0.0.1 down entry only when required

[pini-gh]

See #1609 (comment).

Additionally, fix a warning due to a missing backslash escaping from previous merge request.

[buchdag]

Do you think the test cases could be expanded to include the following:

• multiple (three ?) containers with the same VIRTUAL_HOST, one on an unreachable network: assert that we still don't get server 127.0.0.1 down in this setup
• single container with multiple unreachable networks (I don't know if it's possible): assert that we only get server 127.0.0.1 down once

Or maybe it's overkill ?

Excellent work anyway, thank you.

[pini-gh]

* multiple (three ?) containers with the same `VIRTUAL_HOST`, one on an unreachable network: assert that we still don't get `server 127.0.0.1 down` in this setup

This implies that they use the same VIRTUAL_PORT, doesn't it?

* single container with multiple unreachable networks (I don't know if it's possible): assert that we only get `server 127.0.0.1 down` once

Will give it a try.

[buchdag]

This implies that they use the same VIRTUAL_PORT, doesn't it?

Not necessarily. The idea was to cover the (undocumented, admittedly) use case of people doing load balancing behind nginx-proxy as described in #1654, but failing this test won't mean that this use case would stop working. Both this test and the other one are more unit testing than functional testing, hence the overkill question.

I have no problem merging this now and adding further tests later of you agree that they'd be nice to have.

[pini-gh]

single container with multiple unreachable networks (I don't know if it's possible): assert that we only get server 127.0.0.1 down once

@buchdag I don't know how to do it using docker-compose v1 syntax. And on my box test_composev2 fails because test_nginx-proxy_1 and test_web_1 are on the auto allocated test_default network while the nginx-proxy-tester instance is on the bridge network. Then the tester can't talk to the usecase instances. Any idea?