Backend Architecture
Putting it together: free-recall review
Retrieval beats re-reading. For each prompt, reconstruct a full answer from memory across the whole track before you open the model answer — the effort of pulling the connections back together is what makes the synthesis stick.
Reconstruct the unit’s spine without looking back: why composition is not addition, how the timeout budget nests, what makes a cascade metastable, why goodput is the target under overload, how RED and percentiles make the system visible, and why readiness is a checklist.
- 01Why is 'composition is not addition' the core claim of the backend track, and where do the hardest bugs live?
- 02Walk one POST /payments request down the stack and name the guard at each layer, then explain why the timeout budget must be nested.
- 03Walk the canonical latency cascade link by link, and explain why the request path and the shutdown path are the same graph.
- 04What is metastability, why does fixing the original cause not fix it, and what must you do instead?
- 05Under overload, why is rejecting requests the correct behaviour, and how is goodput different from throughput?
- 06What does RED measure, why must you watch percentiles instead of the mean, and what makes a readiness checklist externalized expertise rather than bureaucracy?
If you could reconstruct each answer from memory you hold the track’s spine: composition is not addition, so correct parts compose into incorrect wholes; the timeout budget nests down the stack so the innermost slow thing fails first; the cascade goes metastable and must be broken at the feedback loop, not the cause; goodput, not throughput, is the overload target, defended by shedding on purpose; RED and the p99 tail make the composed system visible; and readiness is a blast-radius-calibrated checklist that only means something to someone who understands every gate it lists.