PACELC
Reason about consistency and latency both during partitions and during ordinary healthy operation.
2–5 minute refresher
Network state?
If Partition: Availability or Consistency; Else: Latency or Consistency
30second
refresher
refresher
PACELC says: if there is a Partition, choose Availability or Consistency; Else, choose lower Latency or stronger Consistency. It corrects a common blind spot in CAP discussions: coordination costs exist even when every link is healthy.
What problem does PACELC solve?
CAP explains the forced decision during a network partition. Most requests happen without a partition, yet a globally consistent write may still wait for another region. PACELC makes that everyday performance decision explicit.
Decision guide
Key trade-offs
| Choice | What you gain | What it costs |
|---|---|---|
| PC / EC | Single order during failures and normal operation | Coordination latency and rejected partitioned work |
| PA / EL | Fast local responses and continued operation | Staleness and conflict resolution |
| Mixed by operation | Critical invariants stay strong; feeds stay fast | More contracts to document and test |
What happens if?
Healthy regions are 120 ms apart
There is no partition, but a synchronous quorum adds at least one wide-area round trip. A local-first design can respond faster only by allowing another region to lag.
What happens if?
The inter-region link fails
A consistency-first operation rejects or delays writes without its quorum. An availability-first operation accepts local writes and needs deterministic conflict handling after recovery.
Where this appears
Globally distributed databases expose PACELC through replica placement, commit policy, bounded staleness, and per-request consistency. Product teams express it through which actions may be stale and which invariants require cross-region authority.
Senior interview modeHow does PACELC improve on CAP?Show answer
It retains CAP’s partition decision and adds the normal-case trade-off: without a partition, stronger consistency still requires coordination and therefore competes with latency.