If you are joining 12 tables for operational data, you have a design flaw. That’s a reporting query pattern.
Often temp tables, CTEs etc are needed as an immediate fix with redesign as a long term fix. The query planner simply can’t optimize that in a reasonable time or it’s beyond there scope if what it can optimize. Also their solution to “joining in memory” is common.
My DB query rule of thumb:
If the query is hard for you to understand, it’s hard for the planner to understand.
Default to small fast simple queries and pipeline them together.
Sometimes longer and more complex queries run much, much faster. The users complain about "complexity", but when they see the performance comparison they change their minds.
I love these! “Not Quite the Same” sounds a lot like “The Wrong Abstraction” from my list. It’s a really important point and I’m glad to see there are more articles about it!
Yep, I have the wrong abstraction on my list as well. There's a couple on ratfactor that really hit the same point home.
I think the lwn article on "the midlayer mistake" is somewhat the same point but to a different extreme, and in some ways libraries not frameworks too.
"I must not complect.
Complexity is the mind-killer.
Complexity is the little-death that brings obliteration.
I will face complexity and I will permit it to pass over me and through me.
And when it has gone past, I will turn the inner eye to see its path.
Where the complexity has gone, there will be nothing.
Only I will remain."
— Litany Against Complexity
Pleasant to hear about the Object-Relational Impedance Mismatch [1], haven't heard about that concept for a long time - maybe a decade!. The reason I like Dapper [2] is that it makes you use your own sql.
> PostGIS is a great database and its foundational data type is geography.
Hate to "well ackchyually" the author here, but I'm gonna. The foundational data type in PostGIS is not geography, but geometry. geography is geodetic layer on top of geometry. I honestly don't use geography that much because it only supports a subset of the geometry functions.
I personally hate YAGNI, because it so often leads to balkanized APIs that only implement the "needed" features and omit things that a reasonable person would expect because they weren't needed initially. Far better to have a clear and explicable model that is fully and consistently implemented.
When you build a tool (physical or software) or an API (web or library) it should be easy to use. The easier it is to use the higher the probability that people will use it. So, if people build a kind of rudimentary thing and say that is all you need (which may be technically correct), then they do not optimize for likelihood of adoption/usage.
Yes I agree with this. It’s all a matter of taste and interpretation at the end of the day. YAGNI has been useful for me in stopping the development of truly complex featureful work that is not needed. But you can go overkill and strip away everything that makes your application nice to use, or that makes your domain representation syntactically complete. And that’s just stupid.
DDIA is great. I read and re-read the small section about LSM merge tree in the first and second edition of the book and got interested in this stuff. It is always a pleasure to read about internals.
the impact of recursivedoubts i.e htmx creator on software engineering practices and adhering to simplicity is monumental but the industry keeps heading towards complexity.
He links here as some kind of damning proof that ORMs are default bad.
https://openai.com/index/scaling-postgresql/
“It’s a poor craftsmen who blames their tools.”
My ORM rule of thumb: ORM for CRUD not Reports
If you are joining 12 tables for operational data, you have a design flaw. That’s a reporting query pattern.
Often temp tables, CTEs etc are needed as an immediate fix with redesign as a long term fix. The query planner simply can’t optimize that in a reasonable time or it’s beyond there scope if what it can optimize. Also their solution to “joining in memory” is common.
My DB query rule of thumb: If the query is hard for you to understand, it’s hard for the planner to understand.
Default to small fast simple queries and pipeline them together.
Sometimes longer and more complex queries run much, much faster. The users complain about "complexity", but when they see the performance comparison they change their minds.
http://steve-yegge.blogspot.com/2006/03/execution-in-kingdom... https://www.parsonsmatt.org/2017/10/11/type_safety_back_and_... https://lwn.net/Articles/336262/ https://tomasp.net/blog/2015/library-frameworks/ https://ratfactor.com/cards/not-quite-the-same https://matklad.github.io/2023/11/15/push-ifs-up-and-fors-do...
http://steve-yegge.blogspot.com/2006/03/execution-in-kingdom...
https://www.parsonsmatt.org/2017/10/11/type_safety_back_and_...
https://lwn.net/Articles/336262/
https://tomasp.net/blog/2015/library-frameworks/
https://ratfactor.com/cards/not-quite-the-same
https://matklad.github.io/2023/11/15/push-ifs-up-and-fors-do...
The last article links here, which I enjoyed:
https://venge.net/graydon/talks/VectorizedInterpretersTalk-2...
I think the lwn article on "the midlayer mistake" is somewhat the same point but to a different extreme, and in some ways libraries not frameworks too.
grug wonder why big brain take hardest problem, factoring system correctly, and introduce network call too
seem very confusing to grug
^ This is the best and funniest paragraph of text that I’ve read this year
> given choice between complexity or one on one against t-rex, grug take t-rex: at least grug see t-rex
That section reminds me of The Litany Against Fear.
Edit "lets" -> "makes".
[1] https://en.wikipedia.org/wiki/Object%E2%80%93relational_impe...
[2] https://github.com/DapperLib/Dapper
Hate to "well ackchyually" the author here, but I'm gonna. The foundational data type in PostGIS is not geography, but geometry. geography is geodetic layer on top of geometry. I honestly don't use geography that much because it only supports a subset of the geometry functions.
When you build a tool (physical or software) or an API (web or library) it should be easy to use. The easier it is to use the higher the probability that people will use it. So, if people build a kind of rudimentary thing and say that is all you need (which may be technically correct), then they do not optimize for likelihood of adoption/usage.
“If we build like Google, we become Google”
I don’t know why a small company would want Google’s engineering problems.