LMDB

LMDB, short for Lightning Memory-Mapped Database, is a memory-mapped key-value store designed for high performance and data integrity. It stores all data in a single on-disk database file that is mapped into a process’s address space, enabling fast, low-latency access through normal memory operations. LMDB provides ACID transactions and supports snapshot-style reads, with read-only transactions that do not block and do not incur major overhead for concurrent access.

Architecturally, LMDB uses a B+ tree–like structure organized for memory-mapped access. Reads occur directly from the mapped memory, while writes are serialized through a single writer transaction, ensuring a consistent commit. Copy-on-write semantics protect data integrity during updates, so readers can proceed while a commit completes. LMDB supports multiple named databases within a single environment and exposes a simple transactional API that is designed to be embedded in applications.

The API is a C library, with core concepts including environments (MDB_env), databases (MDB_dbi), transactions (MDB_txn), and key/value pairs (MDB_val). The API provides both read-write and read-only transactions, and it guarantees durable commits by flushing changes to disk at commit time (subject to configuration options). Bindings exist for many programming languages, making LMDB usable from a wide range of ecosystems.

Key characteristics and usage: LMDB is optimized for read-heavy workloads and offers very low latency for simple key-value access. It is commonly used as an embedded database or cache, suitable for applications requiring strong consistency and fast lookups. Limitations include a single-writer constraint that can bound write throughput and certain size and addressing considerations, which are important to understand for large-scale or highly concurrent write workloads.

Licensing and ecosystem: LMDB is released under a permissive OpenLDAP Public License and has influenced related projects and derivatives, contributing to its broad adoption in performance-critical storage scenarios.