Design globally consistent metadata service

## System Design Prompt Design a **global-scale metadata service** used by many internal products. ### Goals / Requirements - **Writes must be strictly consistent** (treat as *linearizable* writes; no conflicting updates). - **Reads must be low latency** for clients across multiple regions. - Global deployment (multi-region). Assume high QPS and large metadata cardinality. ### What to cover - Data model and API surface (e.g., `Get/Put/List/Watch`, versioning). - Replication and consistency approach; explicitly justify **why not eventual consistency**. - How you handle **network partitions** and **region-level outages**. - **Degradation strategy** under failures (what continues to work, what becomes read-only, etc.). - A **rollback / recovery** approach for bad writes (operator error, buggy deployment). ### Constraints (you may assume) - Metadata items are typically small (KBs), but the keyspace is large. - Clients are latency-sensitive on reads; writes are less frequent but correctness-critical. - You may partition by tenant and/or key.