Scaling down message processing containers for queues

[jtobinisaniceguy]

I have a question for scaling down message processing containers when they are connected to a queue.

Scaling up is simple, when number of messages in a queue exceeds a defined limit, another container is added, however when the number of messages in a queue decreases to below a defined limit, the potential for removing containers can occur.

When this happens, it is possible that, a to be deleted container, may be actively processing a message, which may take significant time and could involve multiple stages with persistent states.
My question is: how does scaling down ensure atomic message processing and avoid partial message processing with potential for message deletion from the queue when the container is deleted?

What customization options are there to query the to be deleted container for activity status and control?