Dragonfly 1.39 is here with major new search capabilities, performance improvements (as always), and a lot more.
Search: `FT.HYBRID` brings native text + vector fusion (LINEAR and RRF strategies) to a Redis-compatible interface, introducing a single command for RAG pipelines,
Dragonfly
1,100 posts
Dragonfly is a drop-in Redis replacement, delivering 25X better performance at 80% lower cost. Host it yourself for free or pay per GB for Dragonfly Cloud.
Joined June 2022
- We're in LA! 🌴 AWS Summit is tomorrow — find us at booth 244 to talk in-memory infrastructure built for extreme throughput at scale. Come say hi! @awscloud #AWSSummits
- Dragonfly Ascent is our first-ever community conference — a virtual event featuring talks from the Dragonfly community. 🎙️ Call for Speakers is now open. Have a story to tell? We want to hear from you. ✨ Submit a talk → hubs.la/Q04k0jsq0
- Dragonfly SSD Data Tiering is GA. Hot data in RAM. Warm and cold data on NVMe. Keys always in memory. Zero application changes. Dragonfly is the only source-available, Redis-compatible in-memory data store with SSD tiering. Redis and Valkey don't have it. In production at
- Dragonfly has been named to the 2026 @Redpoint InfraRed 100, the 100 private companies defining the next era of cloud infrastructure. Multi-threaded. Shared-nothing. 25x the throughput of Redis on the same instance. Thank you to our customers, our community, and Redpoint.
- Teams are replacing up to 10 Redis instances with a single Dragonfly instance using ACL database selectors. Logical isolation + security enforcement. No key prefix conventions. No discipline required from devs. How it works → hubs.la/Q04fNLD60
- A cluster that looks healthy on average can still have one shard quietly running out of memory. You won't see the problem until a hot shard hits its limit and by then you're either dealing with evictions or scrambling to resize. We shipped shard memory balancing in Dragonfly
- Running Redis or Valkey with heavy TTL usage? The expire table is a real cost — a parallel structure you maintain just to track which keys should die. In Dragonfly v1.38, we removed it. Expiry metadata is now embedded directly in each key. On a 10M small-value workload, memory
- Redis was built in a single-threaded era. Modern cloud servers have 64, 96+ cores. Redis uses one of them. At multi-terabyte scale, that mismatch compounds fast: more shards, more nodes, more ops overhead, higher cost. We wrote about what a rethink actually looks like. 👇
- we run a fuzzer on every Dragonfly PR it reads the diff, asks an LLM which edge cases to target, then spends 15 minutes hammering those exact code paths first run caught a real XPENDING off-by-one that unit tests missed full writeup by @VYavdoshenko →
- Vector search just got a serious upgrade in Dragonfly v1.37. 📈 Up to 7x higher throughput ⚡ Up to 65x lower latency If you're running RAG or high-concurrency vector workloads — this one's for you. Read the full announcement:
- 🎉 30,000 stars on GitHub! ⭐ From launch in June 2022 → 10k stars in just 75 days → 30k stars. Massive thanks to our global community of users running Dragonfly in production, contributors shaping the roadmap, and every single stargazer voting with confidence. In 2026,
- Feature explosion in e-commerce machine learning? It’s an infrastructure problem. In our latest blog post, we break down how Dragonfly delivers the sub-millisecond latency and massive throughput modern feature stores demand. #ML #FeatureStore





