In general I would recommend putting some kind of protection between machines running Docker (and especially orchestrators) and the internet. This could be cloud provider mechanisms (security groups, ACLs, etc), a firewall appliance, NAT gateway configuration, etc. depending on the situation. It's not necessary, but it makes the situation easier to audit/validate, and more layers of protection seldom hurt. If nothing else it means that much of the time you'll need to make a mistake in two places instead of one, in order to have an unintended exposure.
Use nmap to evaluate your policy from the outside, don't try to validate it in your head by inspection.
"The EXPOSE instruction does not actually publish the port. It functions as a type of documentation between the person who builds the image and the person who runs the container, about which ports are intended to be published."
Second paragraph:Although the similar words with different meanings no doubt contribute to confusion.