Mechanics of rolling-update: is it robust?

We are in the process of building a rancher + cattle implementation to manage our containers. Trying to get a sense of how upgrades might work. We will leverage a load balancer to distribute http and tcp traffic across multiple containers for relevant services. If we upgrade, either in-service or rolling, I understand that new contains can ramp up before the old containers are deprecated. It appears the terminating the old instances really means scaling them down. What’s the assurance that existing traffic to the old containers are drained before they are deprecated or de-registered from the load balancer? Have searched around and haven’t found answer. Any reference material would be helpful.

Sounds like you would need this enhancement to be implemented.