Dragonfly

What is Dragonfly?

Dragonfly is a Redis- & Valkey-compatible data store for heavy workloads.

Try freeTalk to us
Redis:Single-threaded.Dragonfly:Multi-threaded.GETSETGETSETCore 0Core 1Core 2Core 3IdleIdleIdleKeyspaceGETSETGETSETCore 0Core 1Core 2Core 3Shard 0Shard 1Shard 2Shard 3
Redis runs one core per instance. Dragonfly runs a thread on every core.

Most teams see the same problems scaling with Redis or Valkey: fragile clusters that crash under load, degraded performance that misses key latency windows, and rising costs even though hardware goes underutilized.

Dragonfly solves these with its multi-threaded engine and thread-per-core model, so it scales vertically and removes the bottlenecks caused by a single-threaded architecture.

What are the common Dragonfly use cases?

Dragonfly is ideal for demanding, real-time applications where a single-threaded data store hits throughput or memory limits.

Common use cases:

Key features of Dragonfly

Dragonfly's engineering improvements build on top of each other to squeeze the maximum performance out of your hardware for the lowest cost.

Multi-threaded engine
Dragonfly runs a separate thread on every core with a shared-nothing model, where Redis and Valkey execute every command on a single main thread. This lets a single node carry far more throughput, so you run fewer nodes for the same workload.
Vertical scaling
Because Dragonfly uses every core, you can grow on a single larger node long before you need to horizontally scale. You avoid the slot migrations, rebalancing, and cluster sprawl that come with scaling Redis or Valkey.
Larger nodes
Fewer, larger nodes (600 GB preconfigured, larger on request) keep more requests on one machine. That cuts cross-node network traffic, and every network hop you remove lowers latency. Cloud providers give larger instances more network bandwidth, and fewer nodes means fewer points of failure.
SSD data tiering
Keys and metadata stay in RAM while large values move to local NVMe. A data store can hold up to 8x more data than its memory, so large datasets cost far less to keep online.
Redis and Valkey compatibility
Dragonfly speaks the Redis and Valkey protocol and supports 308 commands. Existing clients and libraries connect unchanged, so adopting Dragonfly usually takes only a configuration change.

How Dragonfly compares to Redis and Valkey

Redis and Valkey execute commands on one thread per instance, can offload network I/O to extra threads, and scale by adding shards. Dragonfly executes commands on every core and scales vertically first, then out, while keeping protocol compatibility.

Threading
RedisOne main thread executes commands. Optional io-threads handle network reads and writes
ValkeyOne main thread executes commands. I/O threads (8.0+) handle reads, parsing, and writes
DragonflyEvery core executes commands, thread-per-core, shared-nothing architecture
Scaling
RedisShard across more nodes with Redis Cluster
ValkeyShard across more nodes in cluster mode
DragonflyScale one instance vertically past 600 GB and millions of QPS. Scale out only when you need more
API compatibility
RedisBaseline
ValkeyForked from Redis 7.2.4. Most apps run unchanged
DragonflySpeaks the Redis and Valkey protocol with 308 commands. Existing clients connect unchanged
Throughputmemtier GET benchmark
Redis1.52M QPS
Valkey1.98M QPS
Dragonfly11.63M QPS
7.6x Redis
5.9x Valkey
P99 latencymemtier GET benchmark
Redis21.12 ms
Valkey9.98 ms
Dragonfly4.58 ms
78% lower than Redis
54% lower than Valkey
Memory efficiencyRedis testValkey 9.0 test
Redis1.43x Dragonfly’s memory at idle on a 5 GB dataset
Valkey149 bytes per entry, about 1.17x Dragonfly’s 127
Dragonfly30% less memory than Redis
15% less than Valkey 9.0

Redis and Valkey details reflect Redis 8.4 and Valkey 9.0. Throughput and P99 latency come from one memtier GET run (128B values, AWS m7g.8xlarge), with Redis and Valkey using 8 I/O threads and each engine at its own maximum throughput. Memory figures come from two separate tests: idle memory under BGSAVE for Redis, and bytes per entry with 64B values for Valkey 9.0. Results depend on workload, hardware, and configuration, so validate against your own traffic.

Why Instacart, Meesho, redBus, and Abusix moved to Dragonfly

Customers consistently report large reductions in cost and latency. Talk to us to see how Dragonfly can help you.

By migrating from managed Valkey to Dragonfly Cloud we’ve been able to cut our cluster size by 70% while also reducing our average and P99 latencies by 40-50% while further optimizing network cost in ways not possible with our managed Valkey service.

Tristan Fletcher

Tech Lead Manager, Instacart

Read the Instacart case study

Deploy anywhere

Run Dragonfly wherever you like, fully-managed or self-managed, on our cloud or yours, or on-prem.

Dragonfly Cloud
Dragonfly Community Edition

Get started with Dragonfly

A drop-in for your Redis client. Be running in minutes.

Try freeTalk to us