Key-Value Store
Subject: Computer and Technology noun /ˈkiːˌvæl.juː ˌstɔːr/ 1 definition
- nosql-databases
- data-storage
- distributed-systems
- storage-engines
-
Definition (noun)
Entry #236 · UpdatedA Key-Value Store (often abbreviated KV store) is a non-relational database or storage engine that organizes data as an associative collection, pairing each unique key with an arbitrary value such as a string, binary blob, or serialized object. Rather than enforcing a fixed tabular schema or indexing fields inside the payload, it exposes simple operations (
GET,PUT,DELETE) centered on direct key lookups. Because the storage layer generally treats values as opaque data, it trades rich querying capabilities—such as joins or ad-hoc secondary filtering—for predictable low-latency access, operational simplicity, and straightforward horizontal scaling.Common Types
In-Memory Caches and Stores: Keep datasets primarily in RAM, optionally backed by periodic snapshots or append-only logs, to deliver sub-millisecond latency for session state, rate limiting, and caching (e.g., Memcached, Redis).
Embedded Storage Engines: Run as in-process libraries that persist data to local disk using B-trees or log-structured merge-trees, eliminating network overhead for host applications and databases (e.g., RocksDB, LMDB).
Consensus-Backed Coordination Stores: Replicate critical metadata across cluster nodes via Raft or Paxos to provide linearizable consistency, change watches, and distributed locks for orchestration (e.g., etcd, Consul).
Horizontally Partitioned Distributed Stores: Shard massive keyspaces across commodity servers using consistent hashing and replication quorums to maximize availability and throughput at scale (e.g., Amazon DynamoDB, Riak).
Example
An e-commerce API stores active shopping carts in a Redis key-value store using keys formatted as
cart:user:84921mapped to JSON-serialized item lists with a 24-hour expiration (TTL). When a customer refreshes the checkout page, the service performs a singleGET cart:user:84921lookup in under a millisecond instead of executing multi-table SQL joins against the primary relational database.