I have this question as well .. granted there are a few targeted cases where SPs make sense, I generally agree with the headache perspective of SPs - such as: the business logic that inevitably ends up in SPs - how do you go about unit testing that in isolation, and particularly from a behavioral standpoint?
I haven't looked in to this in ... decades, but I seem to remember stored procs executed measurably faster 20+ years ago. That was one reason a project I was on pushed them. And the benchmarks they showed... it wasn't unreasonable. But we have faster networks, more memory, ssd, etc - I'm unsure if speed is a big reason any more. IIRC cached execution plans was a big win back then, but may not be as a win today. I'm thinking about sybase and mssql from the late 90s when I was first getting in to this topic.
But... over the years, I've seen push to sprocs where the logic was never version controlled or subject to the same level of testing the regular app source code was. The DBA folks had total control and were treated as a separate class of engineer than mere 'app' engineers, which caused all sorts of problems...
Maybe they’re not properly called sprocs, but there’s at least one good use for logic stored in the database: data retention. I like CREATE EVENT / pg_cron for trimming append-only tables when the data has aged out.
This is very poorly written and is wrong on basic facts.
So what does a stored procedure get us?
> Absolutely nothing! Well, I mean, headache for one.
... we have to deploy migrations to update our queries, and we have to run diff migration_for_my_sproc migration_for_my_sproc_n to see how things changed
1) You can version sql functions in your repo alongside your code and deploy sql function alongside your db migrations(even in the same transaction).
Have one file per sql function and you can also compute checksums to speed it up if like me you have a repo with 600 stored procedures.
2) With stored procedures you have no need to to db.startTransaction on server when executing multiple statements and wait for db round trips. That's often the biggest reason for preferring stored procedures.
3) I have seen systems in healthcare/finance where different teams have no access to underlying tables and the db only exposes sql procedures. Every-time a procedure is called it also adds a log entry to an audit table.
Databases are a great piece of technology! Learning how to use them properly can have huge payoff in terms of business value generation.
EDIT: not mentioned in this article but people often mention testing difficulties with stored procedures.
You can have normal vitest tests testing your postgres functions with in-memory pglite.
I have seen systems in healthcare/finance where different teams have no access to underlying tables and the db only exposes sql procedures. Every-time a procedure is called it also adds a log entry to an audit table.
For a large system that has many different teams working on it, it is better to have a core API layer with clear ownership, than to not have one. But why would you choose to build that with database stored procedures? If this was built 25+ years ago, then that is all the answer that is needed.
I think the idea is that the stored procedures ARE an Application Programming Interface, as an alternative to a REST API.
(I agree with you that for some use cases, an API consisting of a set of stored procedures is just as good as an API layered on top of the database in another language.)
I am uneasy about databases being used in general how they are today. Nobody writes SQL to read and write records directly. it's all pumped through business logic. Then you write a stored procedure to update x-on-y and the business logic breaks, in a scenario that is impossible for the business logic to produce.
additionally the logic is very far away from the data in many cases, obfuscated through layers of data modelling.
If there was a way to bring these closer, that would be nice.
The ancient wisdom which advocated for stored procedures, which modern developers find distasteful, were really advocating for microservices close to your data, which encapsulated security, business logic, and data persistence so that multiple consumers could share the same data without repeating the logic and code.
The fact that some people write those microservices in PL/SQL and some in JavaScript doesn't change the relevance of the encapsulation.
But PL/SQL is a horrifying language. You have barely any facilities for modularity, encapsulation or composition. Every procedure is just one name in a huge flat name space. Development tools are primitive. The language is absurdly verbose - it was state of the art in the 1970s when COBOL reigned supreme.
our application code and DB access are separately versioned, our sprocs can change right under our feet from aberrant (i.e. extremely rare, insane) DBAs
Your database (and its migrations) should be source controlled along with the rest of your code. Problem solved, now you can take advantage of some of the legitimate benefits stored procedures have to offer!
I don't have any concrete examples at this moment to show the advantages of using stored procedures. So, suppose that we have to update 2 tables A, B if data in table C satisfies some conditions. Then stored procedures allow us to send only 1 request to the db while with the update commands we have to send at least 2 requests.
(In our on-prem SQL server environment) We expose stored procedures to web developers like an API contract to ensure a level of integrity and robustness in how they interact with the database. We find SQL knowledge among web developers these days is a toss up and I'd rather have the DBA design the schema and provide stored procedures.
Just learn how to use Liquibase (or flyway, or... ) in your git repo and CICD pipeline deployments alongside your code.
I did it with production Scala apps over a decade ago. I even built sproc TDD test suites.
Besides CI, what benefits were you getting from writing the logic as a stored procedure?
I have this question as well .. granted there are a few targeted cases where SPs make sense, I generally agree with the headache perspective of SPs - such as: the business logic that inevitably ends up in SPs - how do you go about unit testing that in isolation, and particularly from a behavioral standpoint?
I haven't looked in to this in ... decades, but I seem to remember stored procs executed measurably faster 20+ years ago. That was one reason a project I was on pushed them. And the benchmarks they showed... it wasn't unreasonable. But we have faster networks, more memory, ssd, etc - I'm unsure if speed is a big reason any more. IIRC cached execution plans was a big win back then, but may not be as a win today. I'm thinking about sybase and mssql from the late 90s when I was first getting in to this topic.
But... over the years, I've seen push to sprocs where the logic was never version controlled or subject to the same level of testing the regular app source code was. The DBA folks had total control and were treated as a separate class of engineer than mere 'app' engineers, which caused all sorts of problems...
Maybe they’re not properly called sprocs, but there’s at least one good use for logic stored in the database: data retention. I like CREATE EVENT / pg_cron for trimming append-only tables when the data has aged out.
This is very poorly written and is wrong on basic facts.
1) You can version sql functions in your repo alongside your code and deploy sql function alongside your db migrations(even in the same transaction).
Have one file per sql function and you can also compute checksums to speed it up if like me you have a repo with 600 stored procedures.
2) With stored procedures you have no need to to db.startTransaction on server when executing multiple statements and wait for db round trips. That's often the biggest reason for preferring stored procedures.
3) I have seen systems in healthcare/finance where different teams have no access to underlying tables and the db only exposes sql procedures. Every-time a procedure is called it also adds a log entry to an audit table.
Databases are a great piece of technology! Learning how to use them properly can have huge payoff in terms of business value generation.
EDIT: not mentioned in this article but people often mention testing difficulties with stored procedures.
You can have normal vitest tests testing your postgres functions with in-memory pglite.
For a large system that has many different teams working on it, it is better to have a core API layer with clear ownership, than to not have one. But why would you choose to build that with database stored procedures? If this was built 25+ years ago, then that is all the answer that is needed.
Why wouldn't you use stored procedures for it? For internal teams why is it better to have a API?
I can see usecases in which API could make sense, but it doesn't matter in most cases.
SQL already has authorization/authentication built in. For rate limiting you can use something like planetscale's traffic control.
Because SQL is a beautiful declarative language, and a disgusting imperative language.
You mentioned elsewhere "But PL/SQL is a horrifying language. You have barely any facilities for modularity, encapsulation or composition"
That's not true about PL/SQL. You can modularize/encapsulate and compose multiple sql functions.
Even plain SQL can be composable in some cases via views.
Compared to any modern language, the facilities are primitive.
I think the idea is that the stored procedures ARE an Application Programming Interface, as an alternative to a REST API.
(I agree with you that for some use cases, an API consisting of a set of stored procedures is just as good as an API layered on top of the database in another language.)
I agree. The author doesn’t understand advanced database concepts and demands others stop them.
Stored procedures ensure consistency. You know to add an item using one proc call. Not 20 different SQL calls.
Just say you hate SQL. I do. For some god forsaken reason I got placed in a SQL heavy role a while back and was out within months.
I both understand Postgres is the best solution for most DB use cases and I still reach for Firebase for my personal projects.
I am uneasy about databases being used in general how they are today. Nobody writes SQL to read and write records directly. it's all pumped through business logic. Then you write a stored procedure to update x-on-y and the business logic breaks, in a scenario that is impossible for the business logic to produce.
additionally the logic is very far away from the data in many cases, obfuscated through layers of data modelling.
If there was a way to bring these closer, that would be nice.
I think stored procedures are just strong typed "serverless lambdas" which runs really, really close do your data storage.
Yes, this!
The ancient wisdom which advocated for stored procedures, which modern developers find distasteful, were really advocating for microservices close to your data, which encapsulated security, business logic, and data persistence so that multiple consumers could share the same data without repeating the logic and code.
The fact that some people write those microservices in PL/SQL and some in JavaScript doesn't change the relevance of the encapsulation.
But PL/SQL is a horrifying language. You have barely any facilities for modularity, encapsulation or composition. Every procedure is just one name in a huge flat name space. Development tools are primitive. The language is absurdly verbose - it was state of the art in the 1970s when COBOL reigned supreme.
This is an implementation problem, you can introduce a JIT'd language plus some "container" with "versioned deployments" in modern DB clusters.
Your database (and its migrations) should be source controlled along with the rest of your code. Problem solved, now you can take advantage of some of the legitimate benefits stored procedures have to offer!
I don't have any concrete examples at this moment to show the advantages of using stored procedures. So, suppose that we have to update 2 tables A, B if data in table C satisfies some conditions. Then stored procedures allow us to send only 1 request to the db while with the update commands we have to send at least 2 requests.
Apparently “we” need to learn how to use them properly.
(In our on-prem SQL server environment) We expose stored procedures to web developers like an API contract to ensure a level of integrity and robustness in how they interact with the database. We find SQL knowledge among web developers these days is a toss up and I'd rather have the DBA design the schema and provide stored procedures.