Every record needs an ID, and in a system spread across many machines, generating unique ones is trickier than it sounds. An auto-incrementing counter in one database works until you have several databases, because then something has to coordinate who hands out the next number. Random IDs such as classic UUIDs need no coordination but are long and don’t sort by creation time, which can make database indexes less efficient. Snowflake-style IDs, from Twitter, split the difference: a 64-bit number built from a millisecond timestamp, a machine number, and a per-machine sequence counter, so each machine can mint thousands of unique IDs per millisecond on its own and they still sort roughly by time. The catch is that they rely on reasonably accurate clocks.