Distributed Systems
CAP in practice: free-recall review
Retrieval beats re-reading. For each prompt, say or write a full answer from memory before you open the model answer — the effort of recall is what makes the tradeoffs stick when you are reasoning about a real system under partition.
Reconstruct the unit’s core ideas — the formal C/A definitions, why P is mandatory, the binary partition choice, PACELC’s healthy-state axis, logical partitions, and AP conflict resolution — without looking back at the lesson.
- 01Why is Partition Tolerance not a design-time choice, and what is the only real choice CAP leaves you?
- 02State the formal Gilbert-Lynch definitions of Consistency and Availability, and explain why a 'highly available' marketed database is often actually CP.
- 03What does the PACELC theorem add over CAP, and what does choosing EC cost in the healthy state?
- 04Explain how a GC pause or CPU saturation can cause a 'logical' partition in a CP consensus cluster.
- 05You chose AP. What conflict-resolution tax do you owe, and why is Last-Write-Wins on wall-clock timestamps dangerous?
- 06Why does the quorum rule W+R > N give strong consistency, and how does it interact with the CP/AP choice?
If you could reconstruct each answer from memory, you hold the unit’s spine: P is mandatory, so a partition forces the binary CP-vs-AP choice; Consistency means linearizability and Availability means a non-error response from every non-failing node, which is why many ‘available’ databases are really CP; PACELC adds that strong consistency also costs latency in the healthy state; logical partitions come from slow nodes, not just cut links; AP owes a conflict-resolution tax where LWW silently drops data under clock skew; and W+R > N buys per-key freshness while pinning you toward CP under partition.