CAP theorem
Also called: PACELC
When the network between copies of data fails, a system must either refuse some requests (consistency) or answer with possibly stale data (availability).
Definition
C is linearizability (every read sees the latest write), A is that every working server answers, and P is tolerating network partitions. Partitions happen anyway, so the real choice is C or A during a partition, made per operation: uniqueness, money and inventory usually choose consistency; carts, likes and feeds choose availability. PACELC adds that, without partitions, consistency still costs latency.
In practice
During a link failure, a consensus database's minority side refuses writes while the majority side keeps working.