Consul uses DNS before container's hosts file when resolving hostnames

[aldiyen]

On Mac OS X, Consul doesn't seem to be respecting the container's hosts file when resolving hostnames.

To reproduce:

• Explicitly define a hosts file entry for container (via --add-host argument to run command) for a given hostname (e.g. --add-host=my-laptop:192.168.0.7)
• In the host machine's hosts file, set the same hostname to a different IP address (e.g. 127.0.0.1 my-laptop)
• Configure health check in Consul referencing that hostname (tested with HTTP health check, pointing to something like https://my-laptop:443/healthCheck)
• Observe that the health check fails with error HTTP request failed: Get https://my-laptop:443/healthCheck: dial tcp 127.0.0.1:443: getsockopt: connection refused when

Using docker exec to run ping or curl against this same hostname from within the Consul container resolves to the properly configured IP address. It's only the consul process that's resolving incorrectly. I'm not an expert on Docker, so I'm not totally sure what's going on, but it may be due to the way the consul process is executed via the docker-entrypoint.sh script.

[aldiyen]

After some additional troubleshooting, it seems that this is related to the way Go's netgo DNS resolver works on Linux, and this sort of issue has cropped up in other projects. I believe I've fixed the problem. Will make a PR in a few once I do some more testing.

Similar issues in other projects:

• net: Go DNS resolver does not read /etc/hosts golang/go#22846
• Alpine linux doesn't have nsswitch configuration file gliderlabs/docker-alpine#367
• When resolving the Docker Host /etc/hosts is used only after the DNS server moby/moby#34544
• Docker in Docker (dind) container linking problem on macOS docker/for-mac#1302