Design a KV store with QPS API
Company: Databricks
Role: Software Engineer
Category: System Design
Difficulty: hard
Interview Round: Technical Screen
Design a key-value (KV) store that supports basic read/write operations and also exposes an API to report QPS (queries per second).
### Functional requirements
- Support:
- `Put(key, value)`
- `Get(key) -> value | not_found`
- Provide a QPS query API, e.g.:
- `GetQPS(windowSeconds) -> qps`
- (Clarify if QPS is **system-wide**, **per-node**, or **per-key**; and whether it counts **all requests** or only `Get` / only successful requests.)
### Requirements to clarify (ask the interviewer)
1. **Scope of QPS**: cluster-wide vs single instance; overall vs per-key.
2. **Time window semantics**: last N seconds rolling window (sliding) vs fixed wall-clock buckets.
3. **Accuracy**: exact vs approximate; acceptable error bounds.
4. **Latency/SLA**: expected p99 for `Get`/`Put` and for `GetQPS`.
5. **Durability**: must values survive restarts? required RPO/RTO.
6. **Consistency**: strong vs eventual for reads after writes.
7. **Scale**: key cardinality, value sizes, expected QPS, hot keys.
8. **Multi-tenancy**: separate QPS per tenant/app?
### Non-functional requirements
- High availability and horizontal scalability (if cluster scope).
- QPS API should be fast and not significantly impact `Get/Put` performance.
Quick Answer: This question evaluates a candidate's ability to design scalable, highly available key-value storage systems and to integrate real-time metrics reporting such as QPS, testing competencies in distributed systems architecture, data durability, consistency, and observability.