Cloud Parts

Valkey

The Linux Foundation’s BSD-licensed fork of Redis 7.2.4: an in-memory key-value server that Redis clients can talk to unchanged.

What it is
Key-value stores
Used for
Caching
Type
Open source
Made by
Linux Foundation
License
BSD-3-Clause
Facts checked
6 October 2026
Website Source code

What it is

Valkey keeps its whole dataset in RAM and answers commands over the same protocol as Redis, with the same data types: strings, hashes, lists, sets, sorted sets and streams. Valkey 8 is fully compatible with Redis OSS 7.2.4, so existing clients, tools and data files work as they are. Persistence works the same way too: periodic RDB snapshots, or an append-only file that with the default once-a-second fsync can lose about a second of writes. In cluster mode keys are spread over 16,384 slots, each shard has a primary and replicas, and replication is asynchronous, so a failover can lose writes that were already acknowledged. JSON, Bloom filters, vector and full-text search, and LDAP login come as separate modules, packaged together in valkey-bundle.

Since Valkey 8.0 the main thread hands network reads and writes to I/O threads and spends its own time executing commands, which lifted one server from about 380,000 to 1.2 million requests per second on AWS r7g machines. Valkey 9.0 moves whole slots between nodes atomically during resharding, lets hash fields expire one by one, allows numbered databases in cluster mode, and scales clusters to 2,000 nodes; 9.1 reaches 2.1 million requests per second on one server. Amazon ElastiCache and MemoryDB, Google Cloud Memorystore, Aiven and DigitalOcean all sell it as a managed service, and on AWS Valkey costs less than the Redis OSS engine.

Why teams pick it

  • BSD-3-Clause under the Linux Foundation, with a technical steering committee where no organisation may hold more than a third of the seats
  • A drop-in replacement for Redis OSS 7.2: same commands, protocol, clients and data files, and ElastiCache upgrades a Redis OSS cache to Valkey in place without downtime
  • Cheaper managed hosting on AWS: ElastiCache nodes cost 20% less and serverless 33% less than Redis OSS, and MemoryDB writes the first 10 TB a month free
  • I/O threads use more CPU cores while commands still run one at a time, so a single server reaches 1.2 to 2.1 million requests per second

Watch out for

  • It splits from Redis at 7.2.4: features Redis added later, such as vector sets and the built-in Query Engine and time series of Redis 8, aren’t in Valkey, so apps that use them can’t switch back and forth
  • JSON, Bloom, search and LDAP are separate modules: you load them or run valkey-bundle, and managed services choose which ones they offer (Memorystore supports a subset of commands)
  • The usual in-memory trade-offs remain: data must fit in RAM, the default AOF setting can lose a second of writes, and cluster failover can lose acknowledged writes
  • Microsoft doesn’t run Valkey: Azure Managed Redis is built on Redis Enterprise, so on Azure Valkey means a third party such as Aiven

Alternatives

What all key-value stores have in common →
Amazon DynamoDBAWSAWS’s serverless key-value and document database, with single-digit-millisecond reads at any size.Amazon ElastiCacheAWSAWS’s managed Valkey, Redis OSS and Memcached, as node-based clusters or a serverless cache.Amazon MemoryDBAWSA Valkey- and Redis-compatible database that keeps everything in memory but loses no acknowledged write.FoundationDBOpen sourceAn ordered, distributed key-value store with strictly serializable ACID transactions across the whole cluster, built as a small core that other databases are layered on.MemcachedOpen sourceA simple, multithreaded in-memory key-value cache; clients hash keys across independent servers.RedisOpen sourceAn in-memory data-structure server: the default cache, and also queues, counters and leaderboards.RocksDBOpen sourceThe embedded key-value engine Facebook forked from LevelDB: an LSM tree in a C++ library that many databases and stream processors use as their storage layer.Workers KVCommercialCloudflare’s global key-value store that keeps data in a few central stores and caches it in the data centers where it is read, so reads are fast and writes take up to a minute to show up everywhere.

How it fits with other technologies

Run Valkey 9.1 with I/O threads and try commands added since the fork · bash

docker run -d --name valkey -p 6379:6379 valkey/valkey:9.1 valkey-server --io-threads 4
vk() { docker exec valkey valkey-cli "$@"; }

# Redis commands work unchanged
vk SET session:42 '{"user":17}' EX 1800

# a lock that only its owner can release (DELIFEQ, Valkey 9.0)
vk SET lock:invoice:1042 worker-7 NX PX 30000
vk DELIFEQ lock:invoice:1042 worker-7

# one hash field expires on its own while the rest stays (HSETEX, Valkey 9.0)
vk HSETEX user:17 EX 300 FIELDS 1 otp 482913
vk HSET user:17 name Ada
vk HTTL user:17 FIELDS 2 otp name

# compare-and-set on a string (SET ... IFEQ, Valkey 8.1)
vk SET config:version v1
vk SET config:version v2 IFEQ v1

Valkey or Redis today?#

Both grew from the same code until March 2024. Versions checked on GitHub on 2026-10-06.

ValkeyRedis
LicenceBSD-3-Clause for every versionBSD up to 7.2; RSALv2 or SSPLv1 for 7.4; AGPLv3, RSALv2 or SSPLv1 from 8.0
Who decidesLinux Foundation project run by a technical steering committee; maintainers from Amazon, Google, Oracle, Alibaba, Tencent, Apple, Ericsson and PerconaRedis Ltd., which owns the code; new contributions go through its contributor licence agreement
Latest release9.1.2 (September 2026)8.10.2 (September 2026)
Shared baseFully compatible with Redis OSS 7.2.4Has added commands and types since 7.4 that Valkey doesn’t have
Beyond the core typesJSON, Bloom filters, vector and full-text search, LDAP as separate modules (valkey-bundle)JSON, time series, probabilistic types, the Query Engine and vector sets built into Redis 8
On AWSElastiCache and MemoryDB, priced below Redis OSSElastiCache (Redis OSS up to 7.1) and MemoryDB
On Google CloudMemorystore for Valkey (generally available since April 2025)Memorystore for Redis
On AzureNo Microsoft service; third parties such as AivenAzure Managed Redis, built on Redis Enterprise

Where to run it managed#

On-demand list prices for one node of about 13 GB, checked 2026-10-06 (AWS: US East, N. Virginia; Google Cloud: Iowa). A highly available setup needs at least one replica, so double them.

Amazon ElastiCache for Valkey

$0.1752 an hour

A cache.r7g.large node (13.07 GiB) costs $0.1752 an hour, against $0.219 for Redis OSS. Serverless caches cost $0.084 per GB-hour plus $0.0023 per million ECPUs, with a 100 MB minimum (about $6 a month). Valkey 9.0 and later can add a durable Multi-AZ log.

Amazon MemoryDB for Valkey

$0.2163 an hour

Stores data durably in a Multi-AZ transactional log, so failover and restarts don’t lose it. A db.r7g.large Valkey node costs $0.2163 an hour against $0.309 for Redis OSS, and the first 10 TB written each month are free (then $0.04 per GB).

Google Cloud Memorystore for Valkey

$0.1923 an hour

Shards with up to five replicas each, in Cluster Mode Enabled or Disabled, on Valkey 7.2, 8.0, 9.0 or 9.1. A highmem-medium node (13 GB) costs $0.1923 an hour, billed per second; one- and three-year commitments cut that by 20% and 40%.

Aiven and DigitalOcean

other clouds

Aiven runs managed Valkey on several clouds. DigitalOcean replaced its Managed Redis with Managed Caching for Valkey.

See it in a real stack

No architecture uses Valkey yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2024–2026

For fifteen years Redis’s core was BSD-licensed, and cloud providers sold it as managed services. On 20 March 2024 Redis Ltd. announced that from version 7.4 Redis would be available only under RSALv2 or SSPLv1, saying the largest cloud providers commoditised its investment; they would now need a licence deal to offer new versions.

Eight days later the Linux Foundation announced Valkey, picking up development from Redis 7.2.4 under the BSD licence, backed by AWS, Google Cloud, Oracle, Ericsson and Snap. It was led by former Redis contributors, among them Madelyn Olson of AWS, who had been on Redis’s core team, Ping Xie of Google Cloud and Viktor Söderqvist of Ericsson. The first stable release, 7.2.5 in April 2024, only renamed things. Valkey 8.0 in September brought multithreaded I/O, and within weeks AWS offered Valkey on ElastiCache and MemoryDB at lower prices than Redis OSS.

In May 2025 Redis added the open-source AGPLv3 as a third licence option for Redis 8. By then the two had drifted apart, and Valkey 9.0 in October 2025 shipped features Redis doesn’t have, such as numbered databases in cluster mode.

Why it’s built this way

Fork the last BSD release, 7.2.4
Keeping the BSD licence meant anyone, cloud providers included, could keep using, distributing and selling the software without a licence deal.
A foundation with a seat cap per company
A project held by a foundation rather than one company can’t be relicensed by surprise; the governance rules let no organisation hold more than a third of the steering committee.
Stay compatible with Redis OSS 7.2.4
Users could switch without changing code. Valkey 8.0 made no backwards-incompatible changes to command syntax or responses.
I/O threads around a single command thread
The main thread hands network work to I/O threads and keeps running commands itself, orchestrating every job so that no race conditions occur, while more cores take the I/O load.

Timeline

  1. 2024

    Redis moves to RSALv2/SSPLv1; the Linux Foundation forks Valkey (March)

  2. 2024

    First stable release 7.2.5 (April); Valkey 8.0 with multithreaded I/O (September)

  3. 2024

    ElastiCache and MemoryDB add Valkey, priced below Redis OSS (October)

  4. 2025

    Memorystore for Valkey generally available (April); Redis 8 adds AGPLv3 (May)

  5. 2025

    Valkey 9.0: atomic slot migration and per-field hash expiry (October)

  6. 2026

    Valkey 9.1: 2.1 million requests per second on one server (May)

Sources35 · facts checked 6 October 2026

Improve this page on GitHub