Dragonfly

Your Redis cluster isn't a database. It's a compromise. Every shard you've added is a workaround for the same fundamental problem: Redis is single-threaded, and modern infrastructure isn't. At multi-terabyte scale, the bill for that architectural debt comes due. More nodes. More replication. More ops. And you're still leaving 95% of your compute sitting idle. Organizations migrating large Redis deployments to Dragonfly typically see large cost reductions: 20 to 30% cost reduction as a baseline 40 to 60% cost reduction for heavily sharded environments Up to 80% cost reduction when clusters have become significantly over-provisioned over time If your Redis costs keep climbing and your cluster keeps growing, it might be time for a rethink. https://hubs.la/Q0485Qkn0 #DragonflyDB #Redis #InfrastructureEngineering #CloudCost

Most database bugs don't look like bugs until they're in production. They live in the edge cases — the command with one argument instead of two, the ZADD with a NaN score, the MONITOR wrapped inside MULTI/EXEC. Unit tests don't find them. Code review doesn't find them. They wait. This is why we built a self-fuzzing CI pipeline into Dragonfly. Here's what it does: Every night at 2 AM, AFL++ builds Dragonfly with instrumentation and hammers both RESP and Memcached protocol surfaces for 30–60 minutes. But the more interesting part is what happens on every pull request. When a PR touches C++ code, a script sends the diff to an LLM, which reads the changed files and generates targeted seed inputs — commands, argument combinations, and edge cases that specifically exercise the modified code paths. The fuzzer then biases 70% of its mutations toward those targets. The first run caught a bug in XPENDING: arity declared as -2, but the handler unconditionally accessed the second argument. Classic off-by-one. Two-line fix. Without the fuzzer, it would have shipped. A few things we learned building this: → Byte-level fuzzing isn't enough. You need protocol-aware mutators that can generate semantically valid but logically broken inputs. → Persistent mode is non-negotiable. Without it, fuzzing a database runs at ~1 exec/sec. With it: thousands. → LLMs and fuzzers pair naturally — not because LLMs are magic, but because they're good at reading a diff and guessing which edge cases matter. → If it's not in CI, it doesn't exist. Full writeup by Volodymyr Yavdoshenkoon the Dragonfly blog. Link in the comments. #databases #opensource #softtesting #fuzzing #C++

Vector search just got significantly faster in Dragonfly v1.37. In our latest release, we’ve made improvements to deliver 7x higher throughput and 65x lower latency for vector search workloads . Whether you're building RAG pipelines, recommendation engines, semantic search, fraud detection systems, or other high-concurrency vector workloads, this release delivers meaningful performance gains without additional infrastructure. Read the full announcement: https://hubs.la/Q045FZCF0 #vectorsearch #RAG

🎉 We’ve just surpassed 30,000 stars on GitHub! ⭐ We’re incredibly grateful to everyone who helped us get here. Launched in June 2022, Dragonfly reached 10,000 stars in 75 days and has since grown into a global community using it in production, contributing code, and shaping what comes next. Every star is a vote of confidence, and we’re grateful for each one. Looking ahead to 2026, we’re sharpening our focus on what Dragonfly already does best: powering real-time, large-scale, intelligent workloads, especially caching and revenue-critical ML systems where latency predictability, scale, and simplicity are essential. Thank you for being part of the journey and for helping push Dragonfly forward. Join the momentum: https://hubs.ly/Q043RB820

Build a machine learning feature store that won't buckle under scale. Learn how Dragonfly's high-performance architecture is built to handle the explosive growth of features, low-latency retrieval, and concurrent access needs of modern AI/ML and data teams. We'll cover: - The scaling challenge of modern ML feature stores - Why Dragonfly is the optimal foundation - Key design patterns & a live implementation demo

Learn how Akuity boosted Argo CD performance and simplified infrastructure by moving from Redis to Dragonfly. A great example of how API-compatible, modern data infrastructure can drive efficiency without code changes. Check out the full case study to see the results. https://hubs.la/Q041Lw7L0 #ArgoCD Argo Project #GitOps #Kubernetes

Join this session for a complete overview and demo of Dragonfly Cloud, the fully hosted version of the Dragonfly in-memory data store. We will cover: - Dragonfly architecture overview and performance benchmarks - Use cases and results from real customers - A hands-on demo

The 𝘀𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗺𝗮𝘀𝘁𝗲𝗿 is no longer a back-office system. It’s a real-time engine for trading and risk management. In our latest technical post, we explore how to architect a modern security master that scales, handles millions of complex financial instruments, and powers queries from precise filters to semantic search, all with predictable, low-latency performance. Discover how Dragonfly’s architectural design and rich data types & features can transform your financial data infrastructure. https://hubs.la/Q03-Jtw90 #Database #FinancialData #SecurityMaster #FinTech

Building with LlamaIndex? Get one backend data store for chat, docs, vectors, and metadata. 🧠 Dragonfly is a #Redis-compatible powerhouse for AI apps, delivering much higher throughput and simpler system architecture. Read our first blog post of 2026: https://hubs.la/Q03ZJ3Gm0 #AgenticAI #DataInfra

🚀 2025 was HUGE for Dragonfly. Check out our year in review! ✅ 29.5K+ GitHub stars ✅ Dragonfly Swarm (limitless scale!) ✅ Dragonfly Cloud Enterprise (BYOC & guaranteed cost savings) ✅ Events from #KubeCon to #AWSreInvent Grateful to our amazing team and community. Here’s to 2026! https://hubs.la/Q03YD1wF0 #DataInfrastructure #OpenSource #YearInReview