Dragonfly In-Memory Data Store

Dragonfly is a modern, high-performance, drop-in replacement for Redis and Memcached, fully compatible with Redis APIs while delivering significantly better throughput on multi-core systems. Developed by Roman Gershman and team (headquartered in Tel Aviv, Israel) and released as open source in 2022, Dragonfly addresses Redis's single-threaded performance limitations. Key features: multi-threaded shared-nothing architecture utilizing all CPU cores efficiently, with each thread (shard) owning its own keys and connections, avoiding global locks and achieving near-linear scaling. Up to 30x higher throughput than Redis on 32-core machines for GET and SET, processing millions of ops/sec. Redis API compatibility: supports over 180 Redis commands across all data types (strings, lists, hashes, sets, sorted sets, streams, bitmaps, hyperloglog, geospatial). Memcached protocol support. Memory efficiency: up to 30% less memory via specialized dashtable allocator avoiding per-key SDS overhead. Snapshotting: efficient point-in-time snapshots using copy-on-write without blocking writes, snapshotting shards in parallel. Primary-replica replication with full and partial resync. ACL with user management and command-level permissions. Pub/Sub and sharded Pub/Sub. Keyspace notifications. Lua scripting via LuaJIT. JSON module with JSONPath queries. Cluster mode support. TLS for encrypted connections. Fully managed Dragonfly Cloud. Open source (BSL, transitioning to Apache-2.0).