ActiveRecord is awesome! However, it’s not without a few downsides and common pitfalls. The simplicity sometimes comes at a cost of efficiency. A great example is the n+1 problem when scoping an ActiveRecord relation by a belongs_to association.
For example, take the following associated Models
class Employee < ActiveRecord::Base belongs_to :company end class Company < ActiveRecord::Base has_many :employees has_many :offices end
If you were using an Employee instance method to find company offices for the Employee, you may do something like this:
Or you could say:
Both examples will give you the following query:
SELECT * FROM offices WHERE company_id = ?
However, the first example will have to load the Company from the database, instead of just referencing the company_id attribute. The second example is only referencing attributes of the current object and doesn’t need to know about the associated instance of Company.
I ran across this problem when investigating transaction performance on some background processes. Using New Relic, I was able to dig into a specific time frame and see various metrics on all the queries being performed.
You may be thinking, “What about has_many through?!”. It’s a valid point and it avoids the n+1, however, this is not always possible. When working with an Engine or gem, the association may not be defined.
This can be pretty helpful when you are running millions of queries a day. In my case, it cut query throughput on one table from 30k cpm to 20k cpm, that’s a 30% reduction with very minor code changes! This in turn increased throughput on the ‘meaty’ transactions that were actually processing data.