By ByteBot
Valkey 9.1 dropped on May 19 with a change that costs you nothing to deploy: per-key memory usage falls by up to 10% the moment you upgrade — no config changes, no data migrations, no tuning. For services running millions of small keys — session tokens, feature flags, rate-limit counters — that’s real infrastructure savings handed to you for free. And that’s only the start of what 9.1 brings.
If you’re still running Redis 7.2 — which hit end-of-life in February 2026 — or paying Redis Enterprise rates while the community forked and moved on, 9.1 is another data point that Valkey has outgrown its “early fork” label.
What’s New in Valkey 9.1
Memory Efficiency That Actually Shows Up in Production
The headline improvement is memory. Internal pointer optimizations cut storage for strings under 128 bytes by up to 20%. Skiplist restructuring reduces sorted set memory by up to 10%. The combined effect is a reduction of up to 10% per-key for common workloads. This is not a tuning tip — it’s an automatic improvement that kicks in on upgrade. At the scale where Valkey earns its place (millions of keys, multi-gigabyte datasets), this is not a footnote.
Database-Level ACLs Close the Multi-Tenant Gap
Before 9.1, ACL rules in Valkey applied globally — a user either had permission to run a command or they didn’t, regardless of which database they were accessing. The workaround for multi-tenant isolation was running separate Valkey instances per tenant. That’s operationally expensive and wastes memory.
Valkey 9.1 introduces numbered database-level access control. You can now restrict users to specific databases within a single instance — tenant A gets DB 0, tenant B gets DB 1, and neither can touch the other’s keyspace at the ACL layer. This is the feature that makes Valkey viable for multi-tenant SaaS without the instance-per-tenant tax.
TLS Certificate Rotation Without Downtime
Certificate rotation in earlier versions required restarting the Valkey process — a painful requirement in any environment where availability matters. Version 9.1 adds automated background TLS certificate reloading. Rotate your cert, Valkey picks it up without interruption. The INFO command also now surfaces certificate expiration dates, giving operators visibility before certificates go stale. If you run Valkey in any security-conscious or regulated environment, this is worth upgrading for on its own.
CLUSTERSCAN: One Command Instead of Many
Cluster operators will recognize the pain this feature solves. Scanning all keys in a Valkey cluster previously required running SCAN on each node independently and merging the results client-side — fragile, verbose, and easy to get wrong at scale. The new CLUSTERSCAN command provides a unified interface for iterating over keys across the entire cluster in a single operation:
CLUSTERSCAN 0 MATCH "session:*" COUNT 100It’s a small surface-area feature with meaningful operational quality-of-life impact, especially for teams building internal tooling, running key audits, or orchestrating data migrations.
The Performance Story
Valkey 9.1 benchmarks at 2.1 million requests per second at 512-byte payloads with nine IO threads and a pipeline depth of ten. The redesigned IO threading communication model contributes up to 17% throughput improvement. Stream operations (XRANGE, XREVRANGE) are up to 30% faster, and string GET throughput improves by a similar margin.
Treat benchmark numbers with appropriate skepticism — your workload will differ. But the directional signal is consistent: Valkey continues improving independently of Redis, and version 9.1 represents meaningful progress on production-relevant operations.
Migration Reality Check
The migration story remains compelling. Valkey is wire-protocol-compatible with all Redis 7.2 clients — redis-py, ioredis, Jedis, go-redis all work without modification. RDB and AOF files are binary-compatible, so the standard migration path is: stop Redis, point Valkey at the same data directory, start Valkey. Zero-downtime replication-based migrations are also supported for environments where even brief downtime is unacceptable. The official Valkey migration guide covers both paths in detail.
The one caveat worth stating plainly: if your stack depends on Redis Inc.’s proprietary modules — RediSearch, RedisJSON, RedisGraph, or RedisTimeSeries — those have no direct equivalents in Valkey core. Valkey Search 1.2, released alongside 9.1, now combines full-text search, numeric filtering, tag-based lookup, and vector search in a single community module, closing the gap significantly. Verify your specific module usage before migrating.
Who’s Already Moved
Snap migrated 70% of its Redis fleet to AWS ElastiCache Valkey in the first half of 2026 — cutting caching infrastructure costs from $2.1 million per year to $840,000 per year while maintaining 5 billion daily requests. Aiven migrated 15,000 servers. AWS, Google Cloud, and Oracle OCI all chose Valkey for their managed in-memory database offerings. These aren’t early adopters hedging on a fork — this is where enterprise infrastructure decisions have landed.
The Linux Foundation press release for 9.1 lists over 80 contributors. Valkey is not a two-person fork keeping the lights on. For more background on the cloud provider cost comparison, see our earlier piece on AWS, Google, and Oracle choosing Valkey over Redis.
Should You Migrate?
If you’re on Redis 7.2 (EOL February 2026): yes, and your window for a low-stress migration is narrowing as your installation ages without patches. If you’re on Redis Enterprise paying a premium while AWS and Google now offer managed Valkey 20–33% cheaper: the math is straightforward. If you’re deeply invested in Redis Inc. proprietary modules: check Valkey Search 1.2 and the community module status before committing — the gap has shrunk but it hasn’t disappeared.
Valkey 9.1 is a mature release from a well-resourced open source project. The memory improvements alone justify the upgrade for existing Valkey users. For Redis holdouts, the combination of free performance gains, real enterprise adoption signals, and a managed service price advantage makes it harder to argue for staying put.
