Comment by bob1029
12 hours ago
> I was surprised how easy it is to express pretty complicated game logic in SQL. The game logic is just ~5900 lines of SQL.
I still think HN is taking major naps on the capabilities of contemporary SQL.
There are businesses so complicated that maintaining procedural code over the domain is largely infeasible. Implementing business rules in SQL can decompose the problem in ways that allow for a lot more people to interact with it at the same time.
When I was working in semiconductor manufacturing, we relied very heavily on stored procedures and SQL to operate the factory. Very little operational decision logic existed in code. We had hundreds of users who were inspecting and proposing changes to the same set of procedures. Testing this stuff was trivial because we replicated the prod DB every morning and experimented against live data directly. There was no gap between the information of the business and its logic. Most shops are not ran this way. They treat the database like some CRUD retrieval engine instead of the nexus of both the data and logic.
When people advocate for spending big piles of money with Microsoft, Oracle and IBM, they are generally going for something like the above. They want literally one system the business operates inside of. Spreading a solution across 10+ vendors and tools when you could do with one is borderline negligence depending on your role in the organization.
There are very few things I would want to do less than complicated logic in SQL. At least these days we get the alternative of SpacetimeDB functions https://spacetimedb.com/docs/functions But having everything in stringly typed environment with minimal stdlib in things like mssql? Yeah, there's a reason why it's not a popular pattern.
At my current job I'm in charge of developing our data analytics platform, and with it being so data-driven I was able to implement like 95% of the logic directly into the database largely as functions and stored procedures. The remaining 5% of the application code is mostly Python, which merely acts as a connector from an HTTP gateway to call said functions and render their output as XLSX files. It's amazing how much I was able to do with mostly SQL (I still had to break down and use plpgsql at times to handle the more procedural stuff, but that was the exception).
And now I'm in the process of replacing some of the more intensive data analytical stuff with DuckDB, which has been just an awesome experience.
Long live SQL!
Its always a very interesting architectural question when deciding at what level(s) the business logic should live. There absolutely are valid reasons for some to go in SQL, though I tend to avoid putting the most complex logic there when its really tricky.
When it really gets hairy, or when the business logic keeps changing under my feet, I'll try to find constraints I can put in the db as a final backstop while leaving most of the logic somewhere in the application stack.
I can count on one hand the number of people I've worked with that really know SQL well enough to pick up complex business logic at that layer and work with it easily. I've been mainly in small companies for the last decade, I'm sure at larger orgs there are more data engineers running around that could own it.
I hated debugging and fixing stored procedures for legacy 2000s apps in de 2010s. The SP code wasn't treated like regular code, it wasn't source controlled. Mostly bad memories from having business logic inside the DB instead of having it all inside the code.
I have the same memories from many years of working with stored procedures - although you should certainly have been using source control.
But nothing quite matches the satisfaction of someone in a far off engineering group reporting a problem that prevented them from committing data that would have corrupted the database - because your referential integrity constraints protected the database.
In the age of AI and a massive explosion of code touching the database, maybe it makes even more sense to embed logic and constraints inside the database where there's no way to get around them.
We do this as well. There are some downsides of course, but overall it has worked out well for years. Funny we were told the use of stored procedures is a problem and the reason the product should be retired. You should see the dependency graph of the replacement.
> There are businesses so complicated that maintaining procedural code over the domain is largely infeasible
And those businesses can have all the pain of intertwingling their logic with their data model, too, wheee!
This is a cultural decision, not a technical decision, and unless I'm in the mood for a particular kind of swampy adventure in someone else's land, one I stay away from.
> we replicated the prod DB every morning
You may be storing logic in your database, but most businesses store data in their databases, and the amount of data they store (mostly because they can't tell what's important and what's not) makes copy-pasting their database every morning pretty unrealistic.
Some large telcos are doing it. Why would replicating a database daily be problematic?
Probably because you didn't account for that when architecting the system. Obviously it shouldn't be problematic but this is the real world full of poor technical decisions and mildly incompetent management that we're talking about here.
True. The kind of guarantees you can get when staying within the database can often solve whole categories of problems. But, DX of having non-trivial logic inside Postgres is not great either: I feel that I'm making a significant trade-off. I've also seen interesting languages that compile to SQL, would consider those in some cases.
Makes me think about git-based blob storage