For plain caching, sessions and queues, Redis and Valkey are interchangeable, so you're really deciding on license policy and your bill. Move to Valkey if your company bans AGPL or you run on AWS, where it's cheaper. Stay on Redis 8 if you use its time series, vector or search features. Don't migrate for speed: the benchmarks published so far are vendor-run.
Do you have to leave Redis because of the license?
For most teams, no. It depends on your version and on whether your legal team accepts AGPL.
Redis's own blog (a vendor view, so read its framing with that in mind) describes the history. Redis was BSD-licensed through 7.2, and 7.2.4 was the last BSD release. Valkey is the Linux Foundation's BSD fork of it, started in March 2024.
| Redis version | License |
|---|---|
| Through 7.2 (7.2.4 last) | BSD |
| 7.4 | RSALv2 or SSPLv1 |
| 8 | RSALv2, SSPLv1 or AGPLv3 |
The 7.4 row is the awkward one: RSALv2 and SSPLv1 are source-available licenses, and companies with a blanket source-available policy can't run them. Redis 8 added AGPLv3 as a third option, which satisfies some of those policies and fails others, because plenty of enterprises block AGPL outright.
Redis's blog also argues that running Redis unmodified as a backend doesn't normally trigger AGPL's network copyleft. That's the vendor's reading, so have counsel confirm it before you rely on it. If counsel says no to AGPL, the choice is made. It's the same shape as the Terraform vs Pulumi decision, and the HCP Terraform alternatives article covers another license-driven exit.
What will it cost on each path?
Each path below has a different catch, and only some of them are about money.
Stay on Redis 8
This is the pick if your policy accepts AGPL and you use Redis's newer built-in types (more on those below). If you're already on Redis 8, you change nothing and migrate nothing. The catch is that your license position now depends on a legal reading.
Stay pinned on Redis 7.2
Teams on 7.2 under BSD can simply stay. The catch is that you're frozen on the last BSD line while the rest of Redis moves on. In our view it's a holding pattern, not a destination.
Move to Valkey, self-hosted
This is the pick for teams that run their own nodes and have a no-AGPL policy. Valkey is BSD, and your clients carry over (details below). The catch is that you own the migration (see the next section) and the patching afterwards, and Valkey won't have Redis 8's newer types natively.
Move to Valkey on ElastiCache
This is the pick if you're already on AWS. AWS's ElastiCache pricing page lists Valkey 20% below Redis OSS on node-based clusters and 33% below on serverless. If you already hold Redis OSS reserved nodes, they carry over to Valkey in the same node family and Region, so switching doesn't strand that commitment. Reservations can cut up to 55% with all upfront payment.
Take an 8 GB serverless cache running steadily at about 2,500 requests per second. Valkey storage at $0.084 per GB-hour is $0.084 x 8 x 730 hours = $490.56 a month, or about $506 once requests are added (Upstash's worked scenario). AWS doesn't list the Redis OSS serverless rate, but a 33% discount implies about $0.084 / 0.67 = $0.125 per GB-hour, which is about $732 a month for storage alone. So the engine switch is worth roughly $240 a month on that cache, and the discount grows with every extra gigabyte.
The catch: the saving only exists on AWS, and it's a list-price difference, not a measured one for your traffic.
Use a serverless service such as Upstash
This is the pick for spiky or small workloads, where per-command billing beats paying for idle capacity. Upstash charges $0.20 per 100K commands, with a free tier of 250 MB and 500K commands a month. Its fixed 10 GB plan is $200 a month for up to 10K commands per second, with replication and failover included. For the 8 GB cache above, that fixed plan costs less than half the ElastiCache serverless bill.
The catch is that per-command pricing can overtake a fixed plan quickly if your traffic is high. Check your command rate before you commit.
Move off Azure Cache for Redis
For Azure teams this isn't optional. Microsoft's retirement FAQ puts the end of the Enterprise tiers at 31 March 2027 and of Basic, Standard and Premium at 30 September 2028, with Azure Managed Redis as the replacement. Since you're migrating anyway, decide your engine and license policy first and let the destination follow.
Prices checked 6 November 2026.
What do you have to change to migrate?
Less than you'd expect. Valkey's migration guide says it's compatible with Redis OSS 7.2 and earlier, speaks RESP2 and RESP3, and works with the same client libraries.
For a standalone self-hosted instance, the cutover is replication:
- Start a Valkey instance and run
REPLICAOFagainst your Redis primary. - Wait until
master_link_statusreportsup. - Repoint your applications at Valkey.
- Run
REPLICAOF NO ONEon Valkey to promote it.
On ElastiCache, AWS documents a one-command cross-engine upgrade from Redis OSS 5.0.6 or later: modify-replication-group --engine valkey --engine-version 9.0. Reads stay available, failover takes a few seconds and the endpoint DNS doesn't change. Rollback is narrow: AWS only supports going back from Valkey 7.2 to Redis OSS 7.1, so test on a copy before you upgrade past Valkey 7.2.
Three things trip teams up:
- Data files from Redis 7.4 and later aren't compatible with Valkey.
INFOreportsredis_version:7.2.4for compatibility, so checkserver_nameandvalkey_versionto see which engine you're on.- Anything using Redis modules needs checking. Run
MODULE LISTon production first.
No source can tell you whether Valkey behaves the same under your load. The benchmarks published so far are vendor-run (Redis, Upstash, Dragonfly), so replicate production traffic to a Valkey replica and compare. The cutover work usually lands on your DevOps engineers, and our guide to DevOps engineer skills and vetting shows what that role covers.
Which features keep you on Redis?
Redis's blog says Redis 8 ships time series, probabilistic structures (Cuckoo filters, Top-K, t-digest, Count-Min sketch), vector sets and hybrid search natively. It says Valkey offers JSON, Bloom filters and search through official modules instead.
If you rely on one of those types in production, moving means rebuilding on modules or something else. If your cache stores strings, hashes and lists, none of it applies, and that's most caches.
What this means for your cache decision
Ask legal one question before your next renewal or license review: is AGPL allowed? If yes and you use Redis-only types, stay on Redis 8. If yes and you don't, you can stay, but price Valkey on your cloud anyway. If no, move to Valkey, and put the Azure dates on the calendar if that's your host.
Then set a budget for the move itself. The engine change is a command; the effort is testing against production traffic, checking modules and agreeing a rollback window. Give that to a named owner with a date, not to whoever has spare time. Our tech stack guides collect the other comparisons we've written, and what each stack costs to staff when the owner has to come from outside.
FAQ
Why Valkey instead of Redis?
Teams pick Valkey for license policy, AWS pricing or both. It's BSD-licensed under the Linux Foundation, and on ElastiCache it costs 20% less on nodes and 33% less on serverless. Speed isn't a reason: the published benchmarks are vendor-run.
Is Valkey compatible with Redis?
Yes, with Redis OSS through 7.2, including RESP2 and RESP3 and the same client libraries. Redis 7.4 and later data files aren't compatible, and Redis-only types need a replacement.
Do I have to leave Redis because of the license?
Not necessarily. Redis through 7.2 is BSD, and Redis 8 offers AGPLv3. You have to move only if your policy bans AGPL and source-available licenses, or if your cloud retires your service, as Azure is doing.
Can I roll back from Valkey to Redis?
On ElastiCache, only from Valkey 7.2 to Redis OSS 7.1. Past that, keep the backup you took before the upgrade and restore from it.
