This makes no sense. Stored procedures are code - just code that executes within your DB, instead of on your app servers. I agree that it makes sense to keep your business logic all in one codebase, but there are times when stored procs are a very efficient solution to business-logic problems.
Overall I agree that this "DB API" solution is not entirely elegant, and might provide as many security holes as it does interesting debugging challenges.
However, there are some situations in which I could see this being incredibly useful. Combining this Nginx's JSON/API modules means some parts of your API could entirely bypass your app servers.
Let's say you have an API that shows the ten items most recently liked by a particular user. Typically, this request would hit Nginx, your app server, then your cache and/or DB. By using the "shared procs API" and some Nginx modules, you could provide the same data by letting Nginx query your cache/DB directly.
There are three common reasons I write UDFs:
* The SQL type system and dependency tracking are deemed helpful. This applies only to 'LANGUAGE SQL' functions, which are just regular queries. Unlike dynamically generated SQL from the application, type checks and dependency tracking on other SQL objects can be employed. This is hard to address in any tooling without having a complete copy of the catalogs (and many databases are not as self-hosted complete as Postgres' catalogs, so sometimes that may not be enough...) and semantic analyzer.
* Diverse language and/or library access. If there is a series of obscure rules to hold simple invariants locked into exactly one application's libraries, it makes it pretty much impossible to use any other library. This can be addressed, but it's often expensive to do so.
* Performance, in rare cases where network round trip dominates execution time and I need more than one plain SQL statement; this is only the case for the integrated 'procedural' languages. One could accomplish something similar by running the application's code closer to the machine, and I think that'd be better, actually, but requires some orchestration.
Another reply to your post said you are making a 'religious argument' - I think there is some truth in that.
If you had said "in my experience, shared procs are not a good place to store business logic", that would have been your opinion, and no-one could argue with you. But you stated, as fact, that "business logic should be written in code, not in data and not in the database". This is not objectively true for all companies, and can be easily disproved. And was disproved, by all the people who said this assumption is incorrect.
Throwing around the names of authors does nothing to help your case (useless "appeal to authority"). I have read some of Robert Martin's work (but none of Martin Fowler's) and still think storing business logic in database rules can be a very good idea.
The web world has produced people for whom there is a 1:1 correlation between app and DB. In the enterprise world, most apps are written to run against existing DBs, some of which have been around for decades. This is scalable development processes. And the funny thing is, web kids will blithely repeat "DRY" without ever understanding the truth of it.
Why?
Your post generally makes a lot of criticisms, but fails to back up any of the points made.
But yes, completely failing to mention a point when attempting to make an argument generally means they fail :)
You might want to read the HN posting guidelines - your comment history suggests you have not done so yet. I have no interest in continuing a conversation with you, based on the level of discourse evident in your previous comments. Have a good day.
https://github.com/gluefinance/PgDeploy
It gives you an extra safety net, in case the version in the database would differ compared to the version in the VCS, even though they shouldn't, someone might have been evil behind your back. :)
PgDeploy lets you preview the change and show you a diff. If the diff is expected, you can proceed and deploy.
You might find find old article interesting, http://ora-00001.blogspot.com/2011/07/mythbusters-stored-pro...
However, I disagree that stored procedures are a bad thing. There are several ways that stored procedures can be used.
1) Named queries (An API around your db!)
2) Set-intensive operations are really best done in SQL. You can do a heck of a lot (of even multi-stage logic!) in a single query when you throw in some in-line views, common table expressions. Writing in stored procedures gives you a single, testable reference for functionality.
3) Exporting application functionality to other programs that may want to use the db (like Excel for example).
4) Unit tests in the db are very, very helpful and highly underrated. Gotta love the fact that rollback means "none of this ever happened."