so pull the unit and debug it at a bench. this has been the traditional method of working on large scale cluster deployments for decades, why do it any differently for mining? Commie's example is the ideal way of running a large group of mining rigs.
because pulling the unit with two PSU and (insert the number of GPU) is a job for two people hehe.
beside pulling out and putting back in is a waste of time. look at what i said above, i fixed my hanging rig faster than you can pull that bar bell from a rack LOL.