Google Cloud Announces General Availability of Memorystore for Valkey 9.1, Claiming Up to 3x QPS

What happened

Google Cloud announced on its official blog that Memorystore for Valkey 9.1 is now generally available. The post is credited to Ankit Sud, a senior product manager at Google, and Jacob Murphy, a software engineer. According to Google Cloud, this version can reach up to 3x the queries per second of its existing Memorystore for Redis Cluster at microsecond latency, meaning responses arrive within millionths of a second.

What is Valkey? Google Cloud explains in the post that in 2024 Redis Inc. changed the license for Redis. It moved from the permissive BSD open-source license to a dual-license model. In response, Google Cloud and other tech companies backed the creation of Valkey as an open-source alternative governed by the Linux Foundation. Put simply, Valkey is a database that keeps data in memory for extremely fast reads and writes. It is often used as a cache, a quick-access store that saves websites and apps from repeating slower lookups.

Read the full description

Sources are collected, independently checked, then reviewed by Jev.

What changed in Valkey 9.1

Performance: redesigned communication between threads

Threads are separate lines of work that a program runs in parallel. I/O threads handle reading and writing network traffic, while a main thread coordinates the work. According to Google Cloud, earlier versions of Valkey assigned each client connection to an I/O thread in a fixed round-robin fashion, meaning the threads simply took turns. The main thread then had to keep checking a list of pending clients to see which work was done. Valkey 9.1 switches to a lock-free, multi-queue messaging architecture, in which threads pass jobs through queues without waiting on one another. It uses three complementary queues, and idle worker threads pick up tasks on demand. The company also says the new version uses two-phase dynamic scaling. When the main thread's CPU usage exceeds 30%, the first background I/O thread starts automatically. After that, threads are added or removed based on how much work is waiting in the queue, and spare processor cores are parked when idle.

Security and management: finer-grained access control

Access control lists (ACLs) are rules that set which users may run which commands on which data. Google Cloud says Memorystore for Valkey recently gained support for ACLs configured through IAM, Google Cloud's identity and access management system, at no extra charge. A single policy can be applied to multiple clusters (groups of servers that share the data), and the feature includes versioned policy revisions and audit logs. Previously, ACL rules applied to an entire instance, meaning a whole deployed database service. The company says Valkey 9.1 can restrict a user's access to specific numbered databases, for example limiting a test account to non-production data.

Developer commands: scanning, atomic operations and expiration

Valkey stores data under names called keys, and a cluster divides its keys into 16,384 slots spread across servers, or nodes. An atomic operation happens as one indivisible step, so nothing else can interfere partway through.

• CLUSTERSCAN: Google Cloud says scanning keys across a cluster used to require querying nodes one by one. That approach could miss keys, return duplicates or fail if data moved between slots or a node failover happened (a backup node taking over from a failed one). The new command uses a cursor, a bookmark that tracks scan progress, which is aware of how the cluster is laid out. It supports either one sequential scan or parallel scans split across the 16,384 slots.
• HGETDEL: atomically reads and deletes a single field in a hash (a record-like group of fields stored under one key) in a single network round trip. The company cites single-use authentication tokens as an example use.
• MSETEX: sets multiple keys at once with one shared expiration time.
• HSETEX: adds NX and XX conditional flags. NX means set only if the field does not exist, and XX means set only if the field exists.

What it means for everyday readers

Most people will never interact with Valkey directly. However, in-memory databases like it often power login sessions, shopping carts and all kinds of real-time services behind the scenes. Google Cloud says Valkey is especially attractive to organizations scaling AI and microservices that need to serve millions of concurrent users. If the claimed performance gains hold up in real-world use, apps built on such services could keep response times steadier during busy periods.

Open-source licensing is another part of the story. Valkey was created after Redis changed its licensing model. By making this new version generally available, Google Cloud shows that major cloud providers continue to invest in open-source alternatives governed by the Linux Foundation. For company tech teams, database-level access control also helps keep live (production) data separate from test environments, which reduces the risk of mistakes.

Browse the latest news in this topic

• Tech news