Valkey vs Dragonfly: Choosing Your 2026 Cache Strategy

The licensing and compatibility divide

Valkey is a vendor-neutral, open-source continuation of Redis OSS. It maintains compatibility with Redis OSS 7.2 and earlier versions by being a fork of Redis 7.2.4. Major industry players like Amazon Web Services, Google Cloud, Oracle, Ericsson, and Snap Inc. support Valkey to ensure open access to high-performance data stores. I recommend Valkey for most new deployments because it remains a drop-in replacement for Redis 7.2 and earlier versions. Existing Redis client libraries and tools work without code changes. However, Redis Community Edition 7.4 and later produce data files that are not compatible with Valkey. If you are already familiar with the complexities of the Redis ecosystem, you know that migrating from Redis Stack features like Time Series or advanced probabilistic structures such as Cuckoo filters will be difficult because Valkey lacks these specific modules. Since Redis 8.0 adopted the AGPLv3 license in May 2025, legal teams often prefer Valkey to avoid the copyleft obligations that require publishing modifications to network services.

Vertical scaling versus distributed clusters

Dragonfly scales vertically using a shared-nothing, multi-threaded architecture written in C++ where each core owns a slice of the keyspace. This design allows a single Dragonfly instance to reach 3.8 million QPS on an AWS c6gn.16xlarge node. While Dragonfly reaches 3.8 million QPS on a single AWS c6gn.16xlarge node, Valkey 9.0 uses SIMD acceleration for AVX512 and NEON instructions to increase the throughput of pipelined workloads for all users. Valkey 8.0 also reduced memory overhead by 9% to 10% for 16-byte keys and 8-byte or 16-byte values. This improvement came from a dictionary restructuring effort. Valkey 8.0 also introduced asynchronous I/O threading to move connection handling and write-buffer flushing onto dedicated threads. Valkey also replaced the cluster keyslot linked list with a per-slot dictionary to reduce pointer chasing. In contrast, Redis memory usage can spike to three times its normal amount during a BGSAVE operation due to copy-on-write mechanics. Dragonfly avoids these spikes because it uses io_uring for asynchronous I/O. I suggest Dragonfly if you need to replace a sharded Redis Cluster with a single process on one large machine. Dragonfly hits 10 million QPS for SET and 15 million QPS for GET in pipeline mode on the same hardware.

Feature divergence and workload selection

Feature availability determines the choice for many engineering teams. Redis 8 folded JSON, Search, and Time Series into its core, providing up to 87% faster commands and new vector search capabilities. Valkey provides valkey-json and valkey-search as separate BSD-licensed modules, but it does not offer an equivalent for Time Series or the probabilistic structures found in Redis Stack. Dragonfly provides partial support for JSON and Search but lacks the full coverage found in Redis. I find the memory efficiency of Dragonfly compelling for large caches, as it uses 30% to 80% less memory than Redis for the same payload. For users running high-throughput, write-heavy workloads, Valkey leads in batched pipeline request performance. Azure customers face a forced choice because Microsoft will retire Azure Cache for Redis tiers between 2027 and 2028. Managed services for Valkey include AWS ElastiCache, Google Memorystore, and Heroku.

Does the lack of integrated Time Series support in Valkey prevent your team from migrating?