When to Choose CP vs AP

The Key Question
The choice between CP and AP is not about which is better. It is about matching system behavior to business requirements: What happens if users see stale data? What happens if requests fail?
CHOOSE CP WHEN MISTAKES ARE EXPENSIVE
Financial transactions: Bank balances, payments. Duplicate charges violate regulations and cost money.
Inventory and bookings: Concert tickets, hotel rooms. Overselling creates customer service nightmares.
Configuration: Feature flags. Inconsistent configs across servers can break functionality.
CP Cost: During a partition isolating 40% of nodes, those nodes return errors. Overall availability drops from 99.99% to roughly 99.5% during incidents.
CHOOSE AP WHEN SPEED TRUMPS ACCURACY
Social feeds: Users want something in under 200ms. Missing recent posts temporarily is acceptable.
Search and recommendations: Slightly stale results beat nothing. 2 second index lag is invisible.
Analytics: Aggregations lagging by seconds is preferable to blocking writes.
AP Cost: You need conflict resolution. With last-write-wins, 200ms clock skew causes lost updates. At scale, roughly 0.01% of concurrent updates are lost without careful handling.
THE NUMBERS
Payment service at 50K TPS: CP gives 75ms p99 writes, 40% error rate during partitions. AP gives 5ms p99 but may create duplicates. For payments, occasional errors beat complex reconciliation.

💡 Key Takeaways

✓Choose CP when correctness errors are expensive: payments, inventory, bookings. Accept occasional 40% error rate during partitions over data conflicts.

✓Choose AP when speed and availability matter more: social feeds, recommendations, logs. Accept seconds of inconsistency over blocking requests.

✓CP write latency is 50ms to 150ms cross region for consensus versus 1ms to 5ms for AP single node writes. This affects your throughput capacity.

✓AP requires conflict resolution: last write wins loses data with clock skew over 200ms, version vectors need merge logic. CP avoids this complexity.

📌 Interview Tips

1Frame the decision around business impact: "If we show stale data, users see wrong inventory. If we return errors, users retry. Which is worse for this use case?"

2Know specific use cases cold: payments need CP (compliance), social feeds can use AP (speed matters more), inventory needs CP (overselling is expensive).

3Mention the availability math: "CP drops to 99.5% during partitions, AP maintains 99.99%. The 10x difference determines our SLA commitments."

← Back to CAP Theorem Overview