In mid 2000s as a lead HPC sysadmin I accidentally pasted 'shutdown now' into the head node of a 200 CPU HPC cluster, which was 3 weeks into a 12 week simulation job! Fortunately that thing never shut down cleanly and it hung as usual, unable to unmount the NFS share used by all the compute nodes. For the next 9 weeks I was staring at that screen hoping it would keep going and I would not be fired. In the end nobody was ever the wiser.