I'm waiting for the company that creates an AI coding agent as a bootstrap, which they then use to solve real problems and make their trillions.

Where latency is a critical constraint and is smaller than what’s practically possible then forget about global consensus with low latency. You will have to come up with some other strategy. The technique usually it’s fatter pipes for high throughput and eventual consistency only. If you can live with >= 8ms, 99.99% of the apps fall into this, it’s doable.

You are correct that a managed service hides replication complexity from the end user, but then why do people use a managed service? It’s for resilience and HA. Failover is not the same between an eventually consistent replicated MySQL/Postgres and a strongly consistent distributed SQL database like TiDB, where an application won’t even notice a storage node failure. Once you throw scale into the mix, welcome to manual sharding hell. My original point still stands 🙂

Dunno, once you throw in replication and eventual consistency the burden is pushed to the application. My preference would be to write my application as if it’s talking to a single node. Adding additional hosts should not require an application change. Assuming, you care about resilience/HA, if you do t then yes single instance of MySQL should work just fine and is less complex to run.

@jfg956 @MydbopsOfficial @PlanetMysqlComT @PingCAP Thanks

@KintuLabs I wonder if this is true, majority of the people who buy expensive sports cars usually don’t drive them on race tracks or even know how to drive them well. Wonder if it’s just positioning and marketing buy it and they will come.

@mjkhanga @srsarangi I’ve been following him from when he started, his first videos were on distributed systems. I ended up buying his book on modern CPU architecture on one of my trips to India. After watching the videos on the chapters in the book.

I was under the impression that QNX had busted the myth that micro kernel architecture was too slow.

@lemire A big step forward, sadly compiler support is lagging behind.

@shayonj Interesting, do you have any metrics on database size? Number of connections etc? Like a before and after type of thing.

@walkingriver @dhh I’m no expert so I believe this. I was under the impression that as long as you provide the source code it’s fine.

@walkingriver @dhh What was the licensing issue?

I work for PingCAP 🙂 More seriously, I haven’t seen a single business installation where HA/Resiliency was not important. MySQL story here is much better than PG. Scaling I think is usually secondary problem, a good problem to have. Databases like TiDB, there’s Yugabyte and CRDB too, are designed from the ground up to be resilient,you can start very small and scale almost infinitely. I’m convinced that if they were around 20 years ago, the Internet would be running on them.

@MarkCallaghanDB Ken sadly passed away a few years ago. Heikki doesn’t get enough recognition. MySQL’s success is based on replication and InnoDB.

FWIW, InnoDB has its own SQL parser and optimizer, with stored procedures etc. but it’s not used for anything substantial. Very simple use by the FTS code.

@MarkCallaghanDB I remember that, we were really looking forward to your joining the team 🙂. Ken was very excited.

The MySQL optimizer team has been quiet about their achievements (I also know why 😀) and have made substantial progress in fixing the core architecture that they inherited according to my sources. I was chatting with Norvald (lead) at FOSDEM and he told me that the refactoring has now been completed and they are in a position to make significant improvements over plan generation. I haven’t looked at the code but I trust Norvald. Will these or future changes make it into MySQL or remain in Heatwave? That’s an open question.