Isn't Arrow biased towards analytics workloads? Like sibling commenter brings up, I'm not sure what that brings to the table for OLTP.
High performance / ergonomic OLTP in general seems to be somewhat neglected in the last decade or so of DB innovation, because in part I think the needs of the industry have been around mass processing of click/log/event-stream data -- firehose of privacy violation :-).
So "insert quickly then let me analyze later" is becoming a highly polished stone, but I think databases still present a very rough demeanour around transactional workloads and the "hip tools" out there reflect that.
I'm personally all for supplanting SQL but only if what replaces it has a sound relational basis.
As for the Alice book, it's great, but it's quite focused on Datalog. Which I think is awesome and an area of my own interest but not in the mainstream of what people think about in terms of "databases" though I wish it was. Most practitioners in software development unfortunately think of a database as "persistence" and not "data management"... And so the heaps of abuse brought on through ORMs etc.
> So "insert quickly then let me analyze later" is becoming a highly polished stone, but I think databases still present a very rough demeanour around transactional workloads and the "hip tools" out there reflect that.
> I'm personally all for supplanting SQL but only if what replaces it has a sound relational basis.
100% agreed. I think the status quo UX is far more rough than the performance. Unfortunately it seems performance is what drives most investment these days unless you are a producer of hip tools.
> As for the Alice book, it's great, but it's quite focused on Datalog
True, although I posted more for the tour of Relational Algebra fundamentals. I guess the SQL section is probably the most out of date part of the book :)
"True, although I posted more for the tour of Relational Algebra fundamentals. I guess the SQL section is probably the most out of date part of the book :)"
I think a better, more accessible, book for the relational foundations is maybe Chris Date's "Database in Depth: Relational Theory for Practitioners". Though it goes out of its way to avoid dirtying itself with SQL and uses some of his "Tutorial D" concepts instead, it gives a good "practical" look at the relational model that I think "practitioners" would find understandable, though it's not without its eccentricities.
"> I'm personally all for supplanting SQL but only if what replaces it has a sound relational basis.
100% agreed. I think the status quo UX is far more rough than the performance. Unfortunately it seems performance is what drives most investment these days unless you are a producer of hip tools."
I applied for and took a job at RelationalAI last fall because I loved what they were doing with building out a system that was really strong on the relational fundamentals. They have a kind of Datalog-ish language (but better/more practical imho) that is quite nice. The talk that Martin Bravenboer did for the CMU lectures really sold me on it. It was what I was looking for for years, as purely relational alternative to SQL (without.. going off the rails into wonky badly thought through NoSQL network-hierarchical-graph land.)
Couldn't make the job work for personal reasons, but was very promising at first.
Unfortunately their direction seems to have gone elsewhere, basically becoming a graph analytics plugin inside someone else's SQL DB: https://relational.ai/blog/pr-snowflake-summit -- I understand why the path, but it's less exciting.
Anyways it's interesting to see how people talk about PostgreSQL and SQL generally on this forum. Like, as if we can't do better. But if someone tries to do better they go off making something without understanding relational foundations and you end up with something unsound like Redis or MongoDB.
High performance / ergonomic OLTP in general seems to be somewhat neglected in the last decade or so of DB innovation, because in part I think the needs of the industry have been around mass processing of click/log/event-stream data -- firehose of privacy violation :-).
So "insert quickly then let me analyze later" is becoming a highly polished stone, but I think databases still present a very rough demeanour around transactional workloads and the "hip tools" out there reflect that.
I'm personally all for supplanting SQL but only if what replaces it has a sound relational basis.
As for the Alice book, it's great, but it's quite focused on Datalog. Which I think is awesome and an area of my own interest but not in the mainstream of what people think about in terms of "databases" though I wish it was. Most practitioners in software development unfortunately think of a database as "persistence" and not "data management"... And so the heaps of abuse brought on through ORMs etc.