ActiveRecord Cannot Read Minds

Engineering Insights

Mark Lodato
#
Min Read
Published On
March 13, 2025
Updated On
March 14, 2025
ActiveRecord Cannot Read Minds

ActiveRecord Cannot Read Minds

TL;DR

  • ActiveRecord has tons of nice, declarative abstractions on top of SQL
  • It can be dangerous to assume what those abstractions produce

Automatic Association Recognition in Where Clauses

ActiveRecord is great at noticing what is meant when trying to filter a query by a record. If
we have the following two models:

class User < ApplicationRecord
  has_many :pets
end

class Pet < ApplicationRecord
  belongs_to :user
end

We can do:

user = User.first
pets = Pet.where(user_id: user.id)

which generates the SQL:

pets.to_sql
# => SELECT "pets".* FROM "pets" WHERE "pets"."user_id" = 1

because user_id is a column of the pets table. But, for cleanliness, we can simplify this to:

user = User.first
pets = Pet.where(user: user)

This correctly generates the same SQL! And, it even works for has_many associations!

pets = Pet.all
pets = User.where(pets: pets)

Which generates SQL like:

pets.to_sql
# => SELECT "users".* FROM "users" WHERE "users"."id" IN (SELECT "pets"."user_id" FROM "pets")

That's great! ActiveRecord took the words describing what we wanted and converted it to SQL for us! But does this always work the way we expect? Let's consider this setup of users
that have animals with names:

User.create(first_name: "Hank").tap do |hank|
  hank.pets.create(breed: "cat", name: "Nala")
  hank.pets.create(breed: "dog", name: "Max")
end

User.create(first_name: "Bob").tap do |bob|
  bob.pets.create(breed: "dog", name: "Nala")
  bob.pets.create(breed: "cat", name: "Felix")
end

Let's say we want to know which users own pets named "Nala" and what breed those "Nala"s are:

nalas = Pet.where(name: "Nala")
User
  .joins(:pets)
  .where(pets: nalas)
  .pluck("users.first_name", "pets.breed")

And one might expect the output to look like:

[["Hank", "cat"], ["Bob", "dog"]]

But instead what we get is:

[["Hank", "dog"], ["Hank", "cat"], ["Bob", "cat"], ["Bob", "dog"]]

What happened? Well, one issue with trying to convert words into SQL is that there are different ways to interpret words. What ActiveRecord got was:

  1. Get users that have at least one pet whose name is "Nala" (Hank and Bob)
  2. Join them to pets to get all pets for both users
  3. Print the name of each person and the breed of each pet that person has

Which is a perfectly valid interpretation and it's very easy to write tests without sufficient
setup data to catch this (trust me; I've done it!). Instead we could write ActiveRecord like:

nalas = Pet.where(name: "Nala")
User
  .joins(:pets)
  .where(pets: { id: nalas })  .pluck("users.first_name", "pets.breed")

or:

nalas = Pet.where(name: "Nala")
User
  .joins(:pets)
  .merge(nalas)  .pluck("users.first_name", "pets.breed")

Summary

It's important to review the SQL generated by ActiveRecord clauses carefully along with
the words describing the desired data.

Learn more about how The Gnar builds Ruby on Rails applications.

Mark Lodato

Related Insights

See All Articles
Engineering Insights
Why Your AI Coding Agent Keeps Making Bad Decisions (And How to Fix It)

Why Your AI Coding Agent Keeps Making Bad Decisions (And How to Fix It)

AI coding agents making bad decisions? The frustration comes from two fixable problems: assumptions and code quality. Here's how to get consistently good results.
Product Insights
From Dashboards to Decisions: Why Traditional BI Can't Keep Up

From Dashboards to Decisions: Why Traditional BI Can't Keep Up

Stop waiting days for dashboards. Learn how BI2AI uses LLMs and RAG to eliminate the analyst bottleneck and turn complex data into instant executive decisions.
Product Insights
Are Your Legacy Systems Bleeding You Money?

Are Your Legacy Systems Bleeding You Money?

Technical debt now accounts for 40% of IT balance sheets, with companies paying a 10-20% surcharge on every new initiative just to work around existing problems. Meanwhile, organizations with high technical debt deliver new features 25-50% slower than competitors. Features on your six-month roadmap? They're shipping them in three weeks.
Previous
Next
See All Articles