I was just helping out with the network at an event. Worked great in testing, but it failed in production due to unicast flooding the network core. Turns out that some of the PoE Ethernet switches had an insufficiently sized CAM for the deployment combined with STP topology changes reducing the effective size of the CAM by a factor of 10 on the larger switches. Gotta love when packet forwarding goes from O(1) to O(n) and O(n^2)! Debugging that in production is non-trivial as the needle is in such a large haystack of packets so as to be nearly impossible to find in the output of tcpdump and wireshark. The horror... The horror...

First big project I worked on a couple of us sped up the db initialization scripts so we could use a less trivial set of test data to stop this sort of shenanigans.

Things like inserting the test data first and turning on constraints and possibly indexes afterward.