Preserve container IP address OR restart all containers in the same network

[m-khvoinitsky]

There is a problem: sometimes (or always?) updated containers have different IP addresses than prior to update and restart. This causes some compose services to fail because their neighbor containers has their previous IP cached. The example (and some proposed workarounds) are described in #186.
The proposed solution is setting nginx resolver ttl to 1 second is not great because in some cases nginx config is baked into an image and you can't easily change it. And after all, disabling cache may harm performance.
So the questions are:

• What is the best practice to handle the issue?
• Would it be possible to make watchtower and docker preserve previous IP address of the updating container?
• If not, may be it makes sense to add an option to restart all containers in the same network of the updating one?