A modern Redis and Memcached replacement engineered for multi-core servers — delivering 25x more throughput, better cache hit rates, and up to 80% lower memory consumption with full API compatibility.
Dragonfly is a high-performance, in-memory data store built as a drop-in replacement for Redis and Memcached. Created from scratch to exploit modern multi-core hardware, it uses a shared-nothing architecture that partitions keyspace across CPU threads — eliminating mutex contention and allowing throughput to scale linearly with core count. On network-optimized cloud instances, Dragonfly reaches 3.8M+ QPS on a single node, and 10–15M QPS in pipeline mode.
Unlike Redis, which is fundamentally single-threaded and hits a CPU ceiling on large instances, Dragonfly runs all data operations in parallel using fiber-based cooperative multitasking. Each thread owns its shard exclusively and communicates with peers only through message passing — never shared locks. This means a 96-core machine delivers proportionally more throughput where Redis would saturate on a single core.
Dragonfly introduces its own Dashtable data structure — an evolution of extendible hashing from academic research — which provides 30% better memory efficiency than Redis in idle state and eliminates the memory spikes Redis exhibits during BGSAVE snapshots. The adaptive caching mode uses a single unified eviction algorithm tuned for real-world hit rates. Prometheus-compatible metrics are exposed natively on the same port as RESP without any extra configuration.
The project ships a managed cloud offering (Dragonfly Cloud) for teams that want the performance without operational overhead. Self-hosted Dragonfly is licensed under BSL 1.1, which permits production use freely as long as you are not selling it as a managed service. At 30,000+ GitHub stars and releases every few weeks, Dragonfly has established itself as the leading next-generation Redis-compatible datastore.
Dragonfly is written in C++17 and built on helio, an open-source I/O library that provides fiber-based cooperative multitasking per thread. The shared-nothing architecture divides the keyspace into shards, each owned exclusively by one thread — eliminating cross-thread mutex contention entirely. Multi-key atomic operations are implemented using VLL (Virtual Lock Lists) from academic database research, enabling lock-free coordination between shards via message passing rather than spinlocks.
The core hash table is Dragonfly’s own Dashtable implementation, adapted from the Dash paper on scalable hashing for persistent memory. Each Dashtable segment holds 840 key-value slots in 56 regular buckets plus 4 stash overflow buckets, with open-addressing and constant-size segments. This design enables incremental resizing without a stop-the-world rehash and supports stateless dictionary traversal — two properties carried over from Redis but achieved with 30% less memory overhead.
Snapshot persistence uses a novel algorithm that eliminates the fork-based memory spike present in Redis BGSAVE. Because Dragonfly’s data is sharded, each shard can snapshot its portion independently while serving reads and writes, removing the need to copy the entire address space at snapshot time. The result is near-flat memory usage during snapshots even at multi-GB dataset sizes. TLS, ACL-based access control, and Lua 5.4 scripting are also supported natively.
What sets Dragonfly apart architecturally is that vertical scaling is a first-class design goal rather than an afterthought. Adding cores to the server automatically increases throughput proportionally, unlike Redis where adding cores beyond one provides no benefit. The fiber scheduler integrates directly with the Linux io_uring or epoll event loop, minimizing context-switch overhead. This combination of data structure research, academic transaction theory, and modern OS-level I/O gives Dragonfly a rare throughput profile that is both consistent and predictable.
Dragonfly is available as a fully managed cloud service through Dragonfly Cloud, which provides auto-scaling, high availability, point-in-time recovery, and multi-region replication. The self-hosted Community edition is free under BSL 1.1 for internal production use. Organizations that need commercial support, SLA guarantees, or plan to build managed services on top of Dragonfly can contact the team for an Enterprise license.
No Code Platforms · AI Development · Developer Tools
Visual LLM workflow platform with RAG pipelines, agent capabilities, and model management for building production AI applications.
Developer Tools · Game Development · Design Tools
Free, MIT-licensed 2D and 3D game engine with one-click multi-platform export and no royalties.
Developer Tools · Databases · Search
The open-source Postgres development platform that replaces Firebase with authentication, real-time APIs, edge functions, storage, and vector embeddings — all built on PostgreSQL.