The application crashes or gets killed by the kernel if you are doing a long running segmentation. It means 30-60 minutes.
It probably happens because the current undo models do not set a limit on the length of the undo history, just keep adding operations until you clean the whole stack.
Version:
I used 2015.05.2. It is very likely reproducable with the master as well. The relevant code (undo controller and undo models) has not changed.
Steps to reproduce:
Start using the draw tool and check with "top" in a terminal window how the used memory grows. After a long time this would make the application crash.
We saw crashes like this with the MITK segmentation plugin, and also with our segmentation plugin that uses a different draw tool, but the same undo/redo framework.