Valkey 9.2 Launches With Forkless Snapshots
The open-source in-memory datastore is now generally available with snapshot, replication, and caching upgrades.
Updated on Nov. 19, 2026 in Software

Valkey 9.2, an open-source in-memory datastore, is now generally available. It can take snapshots without forking a separate process, and adds stream compression, smaller backup files, and a new Path Hash data type.
Why it matters
Snapshots under the conventional fork-based approach typically require operators to reserve about 50% extra memory, so eliminating the fork reduces infrastructure overhead. The distributed caching feature also targets the cost of running large language model inference on shared GPU fleets.
Project testing reports replication traffic cut by roughly 50% to 65% through continuous stream compression, which costs about 2% to 4% of one CPU core per replica. Backup files are up to 80% smaller with LZ4 compression, and per-member index overhead for large sorted sets fell 36% to 43% in a five-million-member test.
The players
Valkey
Valkey is an open-source, in-memory key-value datastore that originated as a community fork of Redis.
Alibaba Cloud
Alibaba Cloud is the cloud computing arm of Chinese technology conglomerate Alibaba Group.
Multiverse Computing
Multiverse Computing is a company that develops compression and efficiency technologies for artificial intelligence models.
The details
Valkey 9.2 moves the database-writing job for snapshots to a background thread instead of a forked child process. It also switches large sorted sets from skiplists to B+ trees, and adds LMCache-based distributed KV caching for self-managed inference deployments running vLLM and SGLang, which Multiverse Computing reported let it handle 50% more requests on the same GPUs.
Timeline
- November 19, 2026: Valkey 9.2 article published announcing general availability.
The Tech Race
Valkey 9.2's forkless snapshots depart from the fork-based persistence model inherited from Redis, which forces operators to reserve extra memory during snapshots. The release positions the open-source fork as a competitor in the race to lower the infrastructure cost of serving AI inference workloads.
Developers running Valkey can expect smaller backups and lower replication traffic, though the memory savings from forkless snapshots apply most directly to large deployments. Teams running self-managed inference stacks may find the distributed caching feature relevant to GPU utilization and response latency.
The takeaway
Developers evaluating in-memory datastores should weigh snapshot memory overhead alongside replication and backup efficiency when planning capacity. Benchmarks from project testing should be validated against each workload before adopting the new features in production.
Further reading
Explore more software coverage for context on open-source databases and developer tooling.



