Skip to content
Tech

Valkey Explained: The Redis Fork and Its Sorted Set Changes

Valkey is the Linux Foundation-governed Redis fork. Its 9.2 release, out November 19, 2026, rebuilds large sorted sets on a B+ tree to cut memory overhead.

Updated 4 min read4 sources
Rows of stacked rack-mounted computer servers with small blue status lights in a data center.
Rows of rack-mounted servers with blue indicator lights, shown up close.Photo: Victorgrigas / Wikimedia Commons, CC BY-SA 3.0

The short answer

  • Valkey is a fork of Redis that the Linux Foundation governs, and Valkey 9.2 became generally available on November 19, 2026.
  • Large sorted sets now use a B+ tree instead of a skiplist, cutting per-member index overhead by 36–43% in the authors' benchmark.
  • Most of the gain is in memory, not speed; the largest speed figure, 105% for ZADD updates, comes from a pre-merge benchmark.
  • Commands behave the same, but OBJECT ENCODING returns "btree" for large sets, so tooling that checks for "skiplist" needs updating.
  • Before upgrading, check the ZRANGEBYSCORE and ZCOUNT bound changes and downgrade risks in the release notes.

Valkey is a fork of Redis that the Linux Foundation governs. People are searching for it now because Valkey 9.2 became generally available on November 19, 2026, and its new sorted set design matters most to developers who build leaderboards, queues and other ranked data.

Valkey 9.2 was announced as generally available on November 19, 2026, in a Linux Foundation press release issued from San Francisco. The same day, the Valkey project blog published release notes, and Phoronix reported the release and called Valkey the leading fork of Redis.

The headline change for developers is a new storage structure for large sorted sets. A sorted set keeps members in order by a numeric score. Valkey 9.2 replaces the older structure with a B+ tree, which the project says reduces memory overhead and speeds up some commands.

Press coverage also highlights AI caching. The Linux Foundation release describes a separate distributed KV caching capability, built with LMCache and working with vLLM and SGLang. That feature is unrelated to sorted sets, and its vendor figures have not been independently verified.

What Valkey is

A Redis fork under the Linux Foundation

Valkey is a fork of Redis, and Phoronix describes it as the leading one. The Linux Foundation governs the project, according to its press release. Phoronix also reports that the Redict project, another Redis fork, was deprecated in favour of Valkey only days before the 9.2 release, with no further Redict updates planned.

The sources reviewed here do not explain the Redis licence change or why the fork formed, so this article does not cover that history.

Why it exists and who backs it

The sources name several organisations involved. The 9.2 release post was written by two authors who work at Percona. The sorted set deep dive was written by Rain Valentine of AWS and Dragos Andriciuc of Percona. The Linux Foundation press release quotes Madelyn Olson of AWS, listed as a Valkey maintainer, and names Momento, Alibaba Cloud and Tensormesh as contributors or quoted partners.

What the sorted set change is

Skiplist replaced by fbtree

Before 9.2, large sorted sets above a size threshold used a skiplist plus a hashtable. The skiplist stored each element in a separate allocation. The deep dive, dated November 16, 2026, estimates about 37 bytes of overhead per node before the member string.

Valkey 9.2 replaces this with a high-fanout B+ tree variant called fbtree, which has a 61-way fanout. Leaves link to each other, and inner nodes store a small "feature" that helps pick child nodes faster. When members are deleted, the tree compacts sparse leaves, so shrinking sets or frequent score updates give memory back.

What stays the same for developers

According to the deep dive, functional behaviour does not change, so commands behave the same. Small sorted sets are unaffected.

The one visible change: OBJECT ENCODING returns btree

For large sets, OBJECT ENCODING now returns "btree" instead of "skiplist." Any tool that checks for the literal string "skiplist" needs an update.

What it means for developers

Memory savings

The memory figures are the most concrete part of the change. In a benchmark with 5 million members of 20 bytes each, per-member index overhead falls from 50.3 bytes to 28.5 bytes for sequential data, a 43% cut, or to 32.0 bytes for random data, a 36% cut. The deep dive estimates this saves roughly 109 MB in the sequential case and 92 MB in the random case.

These are per-member overhead figures, not total memory use. The author says the relative saving in total memory is largest for sets with small members and shrinks as members grow.

Speed gains

Speed gains vary by command. The deep dive reports the following changes for a 3-million-member set on AWS Graviton3 (64 cores, five repeats):

CommandReported change
ZADD score update+105%
ZREM (50/50 add and remove)+76%
ZCOUNT+27%
ZRANK+19%
ZRANGE (100-element scan)+6%
ZSCORE+2%
ZRANGEBYSCORE+1%
ZPOPMINabout 0%

The authors caution that these benchmarks ran on a pre-merge pull request branch and were not rerun on the merged commit. The Linux Foundation's "up to 105% faster updates" refers to the ZADD benchmark, not a general speedup. Real results will depend on your workload.

Valkey 9.2 also adds forkless snapshots, replication compression and a Zstandard option, according to the release post. In the release's benchmark, LZ4 compression cut replication bandwidth by 52% on a 3-million-key test with under 1% throughput cost. The Linux Foundation press release gives broader vendor estimates, which are not independently verified.

Checks before upgrading

The release post lists behaviour changes that matter even in a minor release:

  • Check any tooling that looks for the "skiplist" encoding string.
  • ZRANGEBYSCORE and ZCOUNT now reject a bare "(" or an empty string as a score bound. Previously these were treated as 0.
  • Active expiration counts toward the dirty counter, so save points may trigger more often.
  • AOF bases written with lz4 or zstd need a version that supports streaming-compressed RDB, so check before any downgrade.

The release post advises not turning on new features until you are confident you will not need to downgrade.

What happens next

As of November 21, 2026, Valkey 9.2 is generally available. The sources do not give dates for later 9.x patch releases or client library updates, so check valkey.io for current status before planning a migration.

Readers who run sorted sets can start with the release post, review the upgrade notes and test the new encoding in a staging environment. Teams that rely on exact encoding strings or score-bound handling should make those changes first. The AI caching features are a separate decision and should be evaluated on their own vendor evidence.

Questions people ask

What is Valkey?

Valkey is a fork of Redis that the Linux Foundation governs. Phoronix describes it as the leading Redis fork, and Valkey 9.2 became generally available on November 19, 2026.

Does the sorted set change break existing commands?

According to the November 16, 2026 deep dive, commands behave the same. The visible difference is that OBJECT ENCODING returns "btree" for large sets. The 9.2 release notes also change how ZRANGEBYSCORE and ZCOUNT handle bare "(" bounds, so check before upgrading.

How much memory does the new structure save?

Per-member index overhead for 5 million 20-byte members falls from 50.3 bytes to between 28.5 and 32.0 bytes, a 36–43% reduction. These are per-member figures, not total memory, and the total saving shrinks as members grow larger.

Is the 105% speed-up a general result?

No. The 105% figure is for ZADD score updates in a benchmark on a 3-million-member set. It ran on a pre-merge branch that the authors did not rerun on the merged commit, and ZPOPMIN was about unchanged.

Sources

  1. Valkey 9.2 offers forkless snapshots, compressed replication, ACL roles, and more — Valkey project blogvalkey.io
  2. From Skiplists to B+ Trees: Making Valkey Sorted Sets More Memory Efficient — Valkey project blogvalkey.io
  3. Valkey 9.2 Lowers Cost of AI Workloads On Existing Infrastructure Alongside New Distributed KV Caching Capability — The Linux Foundationlinuxfoundation.org
  4. Valkey 9.2 Delivers More Performance, Zstd Compression & Lower Memory Overhead — Phoronixphoronix.com

Keep reading