I'm still not entirely sure what the best option is there. Maybe some clever use of network namespaces, with a named pipe to bridge between the "internal" and "external" universes? Just typing up that idea makes me cringe though.
One workaround to this is to have the container host also put in an SNAT rule, so that anything that it forwards to a container would have the source IP address re-written to appear to come from the container host's IP, or the docker0 bridge IP (172.17.0.1/16)
This is also the correct solution for your cluster.
This was an organization that sustained five mines of uptime for decades.
Crazy to see a fallen (or broken up) titan struggle with basic stuff. I mean, basic compared to their heyday.
It's allocated to "DLA Systems Automation Center," a branch of the US military. The addresses are probably used on NIPRNet/SIPRNet, but not publically routed. (Much like 22.0.0.0/8.)
Don't use Kaspersky!
It's likely incompetence, not malice. If they didn't want people using other DNS, and were willing to fuck with ip addresses they don't own to accomplish that, they'd be blackholing google's and opendns's public caching nameservers too.
It might even have been a conscious decision. Even though it's horrible and the people involved in developing the firmware need re-education. The decision probably went like this: we need an internal address to do something. We can't use 10, 172.16, or 192.168 ranges because those might conflict with internal LANs. 1.x is safe because we all know nobody uses them. The correct decision obviously would have been to get at&t corporate to commit to never using some tiny corner of their address space, and use that. Or 127.a.b.c if that works on the OS. Those options are only needed if they really need an extra IP address. They might not need one after all if they designed their firmware better.