Redict is deprecated. The announcement went up on November 17, 2026, and it says Redict 7.3.6 is the final release. That release from October 2025 patched CVE-2025-49844 and three other CVEs. Nobody will patch the next one.
If you run Redict, or you package it for a distro, you have to move. The announcement makes migrating to Valkey sound like a config rename. For at least one person in the Lobsters thread it wasn't: Valkey refused the existing dump.rdb, and the reason is a single integer in a header file. I'll go through why that happens and the one config line that should get you past it. Then come three migration paths (RDB copy, replication, Docker Compose) and what I'd do if you'd rather go back to Redis 8.
Why the Redict project gave up
Redict forked Redis OSS 7.2.4 in March 2024 under LGPL-3.0. The bet was that a copyleft fork would protect the community from the next licence rug-pull. The announcement admits the bet lost on two fronts. Valkey got the momentum, and Redis went back to an OSI licence when Redis 8 added AGPLv3 next to RSALv2 and SSPLv1 in May 2025.
The author, Drew DeVault, recommends Valkey and specifically advises against going back to Redis. The argument is that Redis still requires a contributor licence agreement, and a CLA is what makes a relicence possible in the first place. One commenter in the Lobsters thread credited antirez for the return to open source and hoped Redis had learned from its mistakes, but admitted to being pessimistic. The company already relicensed once, in March 2024, and it still holds a CLA that lets it do it again. For me that's reason enough to stay off Redis.
Valkey vs Redis 8 vs staying on Redict
| Redict 7.3.6 | Valkey 9.1 | Redis 8.x | |
|---|---|---|---|
| Licence | LGPL-3.0 | BSD-3-Clause | AGPLv3 / RSALv2 / SSPLv1 (your pick) |
| Governance | Independent fork, now deprecated | Linux Foundation | Redis Ltd., CLA required |
| Security fixes | None after 7.3.6 | 3 years per minor (policy) | Ongoing (8.4.2 was a security release) |
| Managed hosting | Effectively none | ElastiCache, Memorystore, Aiven, DigitalOcean | Redis Cloud |
| Loads a Redict RDB | Yes | With rdb-version-check relaxed (see below) | Not a supported path |
The managed-hosting row decides it for a lot of teams. When AWS launched ElastiCache for Valkey it priced it 33% below the other engines on serverless and 20% below on node-based clusters, and existing reserved-node discounts carry over. Valkey 9.1 shipped in May 2026 with automatic TLS certificate reloading and database-level ACLs.
There's one case where I'd pick Redis 8: if you want the bundled query engine, JSON and time series modules, and AGPL is fine for your use.
Why a plain container swap to Valkey fails: RDB version 12
The announcement's container advice is to "pull the valkey container instead". In the Lobsters thread, legoktm reports that this failed for them because "the RDB formats were incompatible". Because they only used Redict as a cache, they deleted the dump and moved on.
The cause is in the headers. Redict's src/rdb.h has carried #define RDB_VERSION 12 since the 7.3.0 tag. Valkey's src/rdb.h says:
RDB 11 is the last open-source Redis RDB version, used by Valkey 7.x and 8.x.
RDB 12-79 are reserved for Redis non-compatible RDB formats
Valkey 9 numbers its own format 80 and up. When a stock Valkey sees version 12, it treats the file as foreign and refuses it. Valkey's own migration guide only promises compatibility with Redis OSS 7.2 and earlier, and a Redict dump isn't one of those.
The escape hatch is a config flag. Valkey 8.1 added rdb-version-check (#1604), and #2543, merged for 9.0, extended relaxed mode to foreign formats. In relaxed mode Valkey loads only the types and opcodes it understands and errors out on anything else. Redict 7.3.0's rdb.h defines the same object types (0 through 21) and opcodes that Valkey 9.0 reads, including the cluster-only SLOT_INFO opcode, so a Redict dump shouldn't contain anything relaxed mode has to reject. Per the shipped valkey.conf, the setting "also applies to replication and the RESTORE command", so it covers all three migration paths below. The default is strict.
I haven't seen anyone publicly confirm a full Redict-to-Valkey load in relaxed mode yet, so do it on a copy first.
Step 0: find out what you actually have
redict-cli INFO server | grep -E 'redict_version|redis_version' redict-cli INFO keyspace redict-cli CONFIG GET dir redict-cli CONFIG GET dbfilename redict-cli CONFIG GET appendonly Write down the key counts from INFO keyspace, because you'll compare them at the end. If the instance is a pure cache and you can survive a cold start, skip all of this: stop Redict, start Valkey on an empty data dir, and you're done. That's what legoktm did.
Path 1: RDB copy (short downtime, simplest)
Stop the writers first, then take a fresh snapshot:
redict-cli SAVE cp /var/lib/redict/dump.rdb /var/backups/redict-dump.rdb Move the config over, as the announcement suggests (/etc/redict.conf to /etc/valkey.conf), and add the line from the section above. Stop Redict, then copy the dump into Valkey's data dir (paths vary by distro, so check what CONFIG GET dir told you):
cp /var/lib/redict/dump.rdb /var/lib/valkey/dump.rdb chown valkey:valkey /var/lib/valkey/dump.rdb The config:
# valkey.conf dir /var/lib/valkey dbfilename dump.rdb rdb-version-check relaxed appendonly no Leave appendonly no for the first start. The Valkey migration doc warns that with AOF enabled, the copied RDB won't be imported. Start Valkey and check:
valkey-cli INFO keyspace valkey-cli INFO server | grep -E 'server_name|valkey_version' Valkey still reports redis_version:7.2.4 for client compatibility, so check valkey_version to see what's really running. When the key counts match, run valkey-cli BGSAVE so the data is rewritten in Valkey's own format. If you use AOF, turn it on at runtime with valkey-cli CONFIG SET appendonly yes, which builds the AOF from the data already in memory. Flipping it to yes in the file and restarting can bring Valkey up from an empty AOF instead of your RDB. Run valkey-cli CONFIG REWRITE so appendonly yes is saved to valkey.conf, then edit the file to set rdb-version-check back to strict. That takes effect on the next restart.
Path 2: replication cutover (near-zero downtime)
Run Valkey next to Redict, with rdb-version-check relaxed in its config (plus masterauth if Redict has a password), and point it at Redict:
valkey-cli REPLICAOF redict-host 6379 valkey-cli INFO replication # wait for master_link_status:up Once the replica has caught up, pause writes in the app, check that master_repl_offset matches on both sides, and promote:
valkey-cli REPLICAOF NO ONE Point the clients at Valkey and resume. The write pause only has to cover the offset check and the client switch. The Valkey docs warn that writes which reach the old primary during the switch can be lost, so don't skip the pause.
Path 3: the Docker Compose swap
services: cache: # was: registry.redict.io/redict:7.3.6 image: valkey/valkey:9.1 command: ["valkey-server", "/usr/local/etc/valkey/valkey.conf"] volumes: - ./valkey.conf:/usr/local/etc/valkey/valkey.conf:ro - cache-data:/data volumes: cache-data: Keep the same named volume and the dump comes along. Make sure dir in the config matches the mount path and that valkey.conf has rdb-version-check relaxed for the first boot. Run docker compose exec cache valkey-cli INFO keyspace before you let traffic back in. If you don't add the flag, you get exactly what legoktm hit.
Going back to Redis 8 instead
Redis 8.2's rdb.h also says RDB_VERSION 12. That looks convenient, but the two version 12s were defined independently, and Redict's own compatibility page says switching back from Redict to Redis is not supported. I wouldn't trust a raw file copy here. Use a client-level copy (RIOT or RedisShake, both listed in Valkey's migration doc) against a staging Redis 8 first, then compare key counts. You'll also need to rename redict-cli to redis-cli in your scripts and swap the systemd units back.
The checklist
-
INFO keyspacecounts recorded before you start - Backup copy of
dump.rdbkept outside the data dir -
rdb-version-check relaxedfor the first load only, then back tostrict - AOF off on first boot, then enabled with
CONFIG SET appendonly yesonce the data is loaded - Scripts updated for
valkey-cli(althoughredis-clistill works against Valkey) - Distro packagers: ship a
redicttovalkeytransition package so people get the upgrade instead of quietly running an unpatched server
I'm moving everything to Valkey, and the CLA is the reason I'm not considering Redis. If you think that's the wrong call, I'd like to hear the argument. What are you running in production now: Valkey, Redis 8, or something else like DragonflyDB or KeyDB? And did your Redict data survive the move?
Sources
- Redict is deprecated and the Lobsters discussion
- Redict 7.3.6 release notes
- Redict compatibility with Redis
- Redict
rdb.hat 7.3.0, Valkeyrdb.h, Redis 8.2rdb.h - Valkey migration guide and valkey.conf (9.0)
- Valkey PR #2543: relaxed RDB check for foreign formats
- Valkey release and support policy
- Valkey 9.1 release post
- Redis: now available under AGPLv3 and antirez: Redis is open source again
- Amazon ElastiCache for Valkey announcement