Clock skew
That different servers' clocks show slightly different times, because quartz clocks drift and NTP sync isn't perfect. A sneaky source of bugs in distributed systems that rely on timestamps to order events.
Problem: if node A's clock runs ahead of node B's, an event that actually happened later on B can get a lower timestamp → wrong ordering. A concrete danger: "Last-Write-Wins" conflict resolution can silently discard the newer update if it happened to get a lower timestamp, and time-based locks/leases can expire too early or too late. NTP usually keeps the difference within milliseconds, but it can jump (or go backward!) during adjustments. Mitigations: monotonic clocks for measuring durations, logical/hybrid clocks (Lamport, HLC) for ordering, or explicit uncertainty intervals (Spanner TrueTime). A fundamental reason why "just use the timestamp" is dangerous in distributed systems. Related to HLC and Spanner TrueTime.