Valkey Turns Two: How a Linux Foundation Fork Became the Open-Source Redis Successor

Overview

In March 2024, Redis Inc. abruptly relicensed its flagship in-memory data store, abandoning the permissive BSD 3-clause license in favor of a dual RSALv2/SSPLv1 model that effectively closed the door on cloud-hosted deployments by competing providers. Within days, Amazon Web Services, Google Cloud, Oracle, Alibaba, Ericsson, and others had convened under the Linux Foundation to fork the codebase under a project called Valkey.

Two years on, that emergency fork has matured into a feature-complete, high-performance successor. As reported by TechCrunch, the alliance of cloud giants was notable precisely because “AWS and Google Cloud rarely back an open source fork together” — a signal of how much pressure Redis’s licensing shift applied to the industry. By the close of 2025, Valkey had shipped three major releases, attracted 346 active contributors, and demonstrated the ability to sustain over one billion requests per second at scale.

The Licensing Break and What Followed

For more than a decade, Redis was the de facto open-source standard for in-memory key-value storage, widely deployed in caches, session stores, leaderboards, pub/sub systems, and real-time queues. The March 2024 license change did not remove existing versions from the open-source record, but it meant that no future Redis releases could be freely hosted as a managed service by cloud providers — the dominant delivery mechanism for the technology at enterprise scale.

The Linux Foundation announced Valkey on March 28, 2024, positioning it as a direct continuation of the open-source Redis codebase under the original BSD 3-clause license. The founding contributors brought substantial engineering resources: AWS contributed the initial fork and continues to maintain key modules; Google Cloud donated a high-performance vector similarity search engine; Oracle and Alibaba Cloud have contributed engineering time and test infrastructure.

The project’s investment rationale, published on the project’s own blog, framed the fork not as a competitive move but as a defense of the open-source compact that made Redis ubiquitous in the first place.

Valkey 8.x: Establishing the Foundation

The first year of Valkey’s existence was primarily about proving the fork could stand on its own and begin improving beyond the Redis baseline. Valkey 8.1, released in April 2025 and announced at KubeCon + CloudNativeCon Europe in London, set the tone for the project’s engineering ambitions, according to the Linux Foundation’s announcement and the Valkey 8.1 release blog post.

The 8.1 release delivered measurable gains over its predecessor:

• Up to 20% reduction in memory footprint for typical key-value workloads, achieved through a new hashtable implementation that eliminates the separate dictEntry structure and embeds keys and values directly in the server object — saving roughly 20 bytes per key-value pair without a TTL and roughly 30 bytes for keys with expiration set, according to the Valkey new hash table engineering post
• Up to 20% better performance for encryption in transit when I/O threading is active, as stated in the Linux Foundation press release
• Approximately 300% improvement in TLS connection acceptance rates by offloading TLS negotiation to I/O threads, as documented in the Valkey 8.1 release blog
• 90% reduction in P100 (worst-case) request latency through active memory defragmentation

Three initial modules also shipped alongside 8.1: a JSON document store contributed by AWS, a Bloom filter module for probabilistic membership testing (also from AWS, claiming 90% memory savings over naive implementations), and a vector similarity search engine from Google designed for AI-driven workloads requiring sub-10ms latency at high query-per-second rates.

As Madelyn Olson, co-maintainer of Valkey and a principal engineer at AWS, stated in the release announcement: “Valkey’s rapid modernization highlights the demand for fast, reliable, and secure data processing.”

With over 150 unique contributors to the core project by its one-year anniversary, the community milestone suggested the fork had achieved genuine independent momentum rather than relying solely on corporate stewardship.

Valkey 9.0: The Architectural Leap

If 8.x was about proving parity and closing gaps, Valkey 9.0 — released October 21, 2025 — represented the project’s first major architectural departure from its Redis heritage. According to Phoronix’s coverage and the Linux Foundation’s release announcement, three features defined the release.

Hash field expiration. Redis has always supported key-level TTLs, allowing an entire key to expire after a set duration. Valkey 9.0 introduced expiration at the field level within hash data structures, exposing eleven new commands including HEXPIRE, HTTL, and HPERSIST. This capability is particularly valuable in scenarios where a hash represents an entity (such as a user session or shopping cart) whose individual attributes have different lifespans — previously requiring application-level logic or separate keys.

Atomic slot migration. Cluster resharding in earlier versions moved data key by key, a process prone to transient redirect errors and elevated latency during the migration window. Valkey 9.0 replaced this with atomic slot-level migration using the AOF (Append-Only File) format, ensuring consistent key routing throughout the resharding process and making large cluster rebalancing operations substantially less disruptive.

Multiple databases in cluster mode. Redis Cluster has historically been restricted to a single logical database (database 0), limiting use cases that benefit from namespace separation. Valkey 9.0 extended full cluster support to numbered databases, enabling multi-tenant and multi-application deployments that previously required separate cluster instances.

As InfoQ reported, the performance numbers that accompanied these features were significant: memory prefetching for pipelined commands yielded up to 40% higher throughput; zero-copy responses for large requests added another 20%; Multipath TCP (MPTCP) support reduced latency by 25%; and AVX-512 SIMD optimizations for BITCOUNT and HyperLogLog operations achieved up to 200% higher throughput on compatible hardware.

At scale — clusters of up to 2,000 nodes — Valkey 9.0 demonstrated the ability to exceed one billion requests per second. Madelyn Olson noted at the time: “Valkey 9.0 is a major step forward in that direction with larger and more stable clusters that can more easily scale to meet user demands.”

Community and Ecosystem Trajectory

Valkey’s 2025 year-end review put the contributor count at 346 active contributors over the course of the year — a figure that suggests the project has moved well beyond the initial corporate coalition. The review also described a growing events footprint: an inaugural Keyspace Amsterdam conference, a Keyspace Beijing event that drew more than 1,000 online viewers, and a presence at AWS re:Invent.

The 2026 roadmap, as described in the year-end review, focuses on four areas: simplifying deployment and scaling operations, optimizing performance for modern multi-threaded CPU architectures, deepening integrations with complementary open-source projects, and expanding local user communities globally.

Cloud managed service availability has followed the trajectory of the open-source releases. AWS ElastiCache and Google Cloud Memorystore both now support Valkey as a first-class option alongside (or instead of) Redis. Oracle Cloud Infrastructure also introduced Valkey support in OCI Cache. This managed service availability is significant because it removes one of the primary adoption barriers for enterprise users, who often rely on cloud-managed databases rather than self-hosted deployments.

Adoption and the Path Forward

Third-party research has begun to quantify how much of the Redis installed base is in motion. Survey data cited by Percona and other industry sources indicated that roughly 75% of Redis users were testing, considering, or had already adopted Valkey as of late 2025, with 83% of large enterprises having begun the transition or actively evaluating it. More than three-quarters of those surveyed planned to rely on third-party enterprise support — a dynamic that has created a support ecosystem around Valkey from providers including Percona, Akamai, and the cloud vendors themselves.

What Remains Unknown

The Valkey story is not yet complete, and several questions will shape its next chapter:

• Divergence rate: Valkey and Redis are technically compatible at the protocol level today, but each major release increases the gap between the two codebases. How long full drop-in compatibility is maintained — and how migration tooling keeps pace — remains to be established.
• Module ecosystem maturity: The JSON, Bloom, and vector search modules shipped as beta or early-access components. Their graduation to production stability, and the arrival of additional modules, will significantly affect Valkey’s appeal for workloads beyond basic caching.
• Redis Inc.’s response: Redis has continued development under its new license and has not been standing still. How the two projects’ feature sets compare over a three-to-five year horizon is genuinely uncertain.
• Governance at scale: The Linux Foundation’s stewardship model has worked well so far, but managing a project with hundreds of contributors and competing corporate interests at the scale Valkey appears to be heading toward is a different challenge than the early days of the fork.

Analysis

What Valkey’s first two years demonstrate is that open-source infrastructure projects with sufficiently broad corporate constituencies can survive — and improve upon — a hostile licensing event at the source. The combination of Linux Foundation governance, immediate cloud-provider adoption, and a growing independent contributor base gave the fork structural advantages that most open-source forks lack.

The technical trajectory is notable in its own right. Reaching one billion requests per second in a cluster, shipping hash field expiration (a feature long-requested in the Redis community), and introducing atomic resharding operations all represent genuine engineering progress rather than maintenance of the status quo. The project appears to be investing the resource advantage of its broad backing into capabilities that materially expand what in-memory data stores can do.

Whether Valkey ultimately consolidates the Redis ecosystem or coexists with a commercially licensed Redis in a two-track market remains the central open question. For now, the data suggests the fork has achieved enough momentum to make that question competitive rather than foregone.