I really enjoyed a talk by Harkrishn Patro, which dives into how the clustering system works in Valkey www.youtube.com/watch?v=P6Cb....

That's basically correct. It started out of customer conversations using ElastiCache without really understanding the nuanced durability/consistency tradeoffs. Once we explained it to them, many of them still chose to keep yolo'ing it though.

For cluster mode, we documented the procedure here for Valkey valkey.io/topics/clust..., but nothing has changed with respect to Redis AFAIK here.

I see some mention of using `CLIENT PAUSE`, but there is a safer variant in `FAILOVER` redis.io/docs/latest/... with standalone distributions. It will automatically pause incoming writes, wait for a replica to catch up, then orchestrate the failover.

Awesome! A fun fact about Valkey, is that the name originally came from ValkyrieDB. We change it because only about half of the folks on my team were able to spell Valkyrie.

A lot of people like `valkey-extended`. We are noodling between that valkey-bundle now. Thanks for the input!

Maybe its time to try valkey.io, the modern open source alternative :)

Haha, oops. I'm a contributor to Valkey, so I've worked on it :). I know many obscure things about the engine.

I'm personally more excited about the memory efficiency stuff Valkey built though, you can read about it here: valkey.io/blog/valkey-...

I have too many alerts on mentions to Redis!

Single threaded, both Redis and Valkey are still pretty close. From my testing Valkey is faster but it is a bit use case specific. Multi-threaded there is a bigger lead for Valkey, but Redis is still working on their architecture, so things might change

It uses an approximate LRU. Each key has 3 bytes of LRU information, and when a key needs to get evicted it samples a few random items and picks the one with the earliest LRU timestamp. It saves 13 bytes compared to a linked list. You can read more information here: valkey.io/topics/lru-c....

I think the difference is we feel empowered to innovate. The Valkey community is really excited right now, so there is a lot going on and that is generating a lot of ideas. With Redis, everything took so long to make happen it just made everything feel slow.

You can find various ways to download it or get the container here: valkey.io/download/rel... as well as read the full release notes here: github.com/valkey-io/va....

It significantly reduces memory overhead (~20%) while being faster in many real world scenarios. It required a bunch of tradeoffs, so we would love to get some folks try it out to make sure we aren't regressing before the GA launch.