Topic 11.4
Senior and Staff-Level Questions
In one line
Open-ended design questions where the answer is a structured argument: requirements, sizing, placement, updates, failure behaviour, observability and security.
Think of it like this
An instructor's exam. You're not tested on driving any more, but on teaching and judging when to drive, which road to take, and what to do when something breaks.
Key ideas
- 01
Use the reasoning chain every time: requirement → exact or approximate → which error is tolerable → size and growth → FPR → memory → hashing → deletes → variant → source of truth → updates → rebuilds → distribution → failures → observability → security.
- 02
Numbers anchor credibility: state memory and false-positive traffic explicitly.
Code & diagrams
1. Clarify: n now and in 2 years, QPS, % of lookups for absent keys, deletes?
2. Error budget: FP cost (extra lookup) vs FN cost (must be 0)
3. Size: n, p -> m, k, memory per node / fleet
4. Placement: local, shared (Redis), per-shard, per-region
5. Freshness: outbox -> Kafka -> updater, watermark bypass, rebuild cadence
6. Failure: fail open or closed per component
7. Observe: fill ratio, observed FPR, lag, sampled FN
8. Security: keyed hashes, rate limits, no raw filter exposureExplain it without notes
Design a web crawler URL filter.
Design cache penetration protection.
Design a billion-user membership check.
Design a malware hash lookup.
Design a distributed Bloom filter.
How do you shard a Bloom filter?
How do you update Bloom filters across regions?
How do you prevent false negatives in distributed systems?
How do you monitor Bloom filter correctness?
How do you choose the false positive rate?
What happens if the Bloom filter is overfilled?
When should you use a cuckoo filter instead?
When should you avoid Bloom filters?
How do you combine a Bloom filter, Redis and a database?
How do you combine a Bloom filter with Kafka?
Practice
Pick three staff questions and answer each in five minutes using the skeleton, stating at least two numbers per answer.
Trade-offs
- ↔
At staff level the trade-off discussion is the answer: memory vs FPR, freshness vs complexity, local vs shared, fail open vs closed.
Done when you can
I can answer all fifteen staff questions with numbers and failure behaviour.