When running at Twitter’s scale, components fail all the time. Hard drives have moving parts. Servers have fans that cool them. These moving parts fail at an especially higher rate. At Twitter’s scale, multiple failures occur every day that might need attention from the hardware and software teams. Jeff Dean's presentation from a decade ago lists the probability of some of these events https://www.cs.cornell.edu/projects/ladis2009/talks/dean-keynote-ladis2009.pdf