As a DBMS implementor, I have always assumed that a DB transaction should not be bound to a server side construct (a thread or a client connection, but instead should be an identifier that is transparent and transferrable between clients (subject to auth, of course).
So client thread A can start a tx, get a txId, that can be moved to client thread B which can use it to continue the tx.
Is that something useful for users of DBMSs? It's not hard to do, and I don't see many systems doing it.
@chrisg cool. I’m a big fan of Datalog though, and would rather get the data out asap, and then play with it in my language. You probably thought of cool ways to debug the db-based functions though. Anyway, I’m not smart enough to have an opinion about dbms systems :)
StabilityAI looks like is going under, and if the numbers in this article are any indication, it gives a good idea of how massively (and therefore unsustainably) "AI" is subsidised
Sweet, just added native functions to #Glowdust so now I can write arbitrary functions in #Rust and still call them as if they were defined in the function store.
Small but welcome quality of life improvement. Plus I learned about phf crate and static maps, not a bad day.