Distributed Systems
Distributed capstone: free-recall review
Retrieval beats re-reading. For each prompt, reconstruct a full answer from memory — spanning multiple lessons in the track — before you open the model answer. The effort of recall is what makes the synthesis stick.
Reconstruct the track’s spine without looking back: quorum overlap, fencing against a stale leader, logical clocks, saga compensations, retry budgets, and the idempotency key that ties them all together at the seams.
- 01State the quorum rule R + W > N and explain what each side of it buys you in the order pipeline.
- 02Why can't a stale leader protect the system by checking 'am I still leader?' before writing, and what does protect it?
- 03Why can't you order events across services with wall-clock timestamps, and what do you use instead?
- 04Why is there no distributed ACID transaction across the four services, and what replaces rollback?
- 05Explain retry amplification and the retry budget that bounds it.
- 06Why is idempotency the load-bearing primitive, and what distinguishes a correct key from one that silently fails?
If you reconstructed each answer from memory, you hold the track’s spine: R + W > N gives read-your-write overlap, fencing tokens defend against a paused stale leader because a self-check cannot, logical or bounded-uncertainty clocks order events that wall clocks cannot, sagas replace distributed ACID with compensations, retry budgets bound amplification, and idempotency keyed on business intent makes at-least-once safe across every seam. The capstone insight: each is correct alone — the failures live where they compose.