"That’s why almost ever network service gets run in a loop—restarting immediately upon crashing, hoping the fault was transient. It often is."

Same line of thoughts as https://ferd.ca/the-zen-of-erlang.html

While hardware transients exist (look up Google papers about DRAM failure frequencies in their datacenters), you more probably run into an inconsistent internal software state and restarting may be better and faster than trying to find out which part of the internal state you still can trust and which not.