What are the possible reasons for a Nack to occur in publisher confirm mode, and can a Nack be reliably produced for testing, short of pulling a cable to the disk or other hardware-based actions?
E.g. sending to a non-existing exchange does not lead to a Nack. It leads to a channel close, just like in non-confirm mode.
Btw my rabbit cluster is running on Windows boxes, which might matter, as the file system works quite differently from those in the unix world.
One way to generate Nacks is to
RABBITMQ_MNESIA_BASEto point to a folder on that drive
A test confirmed that that will actually result in a Nack.