Multithreading in Database Systems

Design and implement a thread-safe in-memory key/value store that will be accessed by thousands of concurrent clients. The service must support get(k), put(k,v), and delete(k) operations and guarantee ACID semantics for every single key. You must ensure that:

1. Readers never block other readers (high read concurrency)
2. A writer blocks only other writers to the same key, not readers or writers to other keys
3. No deadlocks can occur regardless of operation ordering
4. The solution scales to millions of keys and tens of thousands of concurrent operations per second

Start with a single-node implementation, then explain how you would extend it to a distributed setting where the dataset is partitioned across many nodes while still preserving the same concurrency guarantees per key. You may use any language; pseudo-code is fine. Be prepared to discuss the trade-offs of your locking strategy, the memory overhead of your metadata, and how you would monitor and tune the system under production load.