How the N+1 query problem can burn your database

Overview

Have you ever watched a movie/series or experience your self going to a hotel where they have room service?, let's put that you go to one of these hotels, where the hotel’s restaurant is at the first floor, and you’re hosteling at the 10th floor. Now, when you are there, you decide to order something to eat for lunch, and imaging that the waitress instead of bringing you all the food with the compliments (drinks, desserts, etc.) at once, bring you every meal, drink, dessert and that, one by one, that will be so inefficient, because the waitress will have to do many runs before bringing you everything you asked for, going back and forth from floor 1 to 10. This is what the N+1 problem is, getting everything you request in many runs.

The ideal will be to carry what you ordered in a bussing cart like this, so the waitress can bring all at once:

In this article will see how this problem look in code and some solutions you can do to avoid this and ensure the best performance for your application.

Time to see some code

To show how the N+1 looks in your code, I’m going to build a simple console application that prints the available menus to order from a restaurant. For this, we’re going to have a database with a meals and drinks table, in the menu, each meal will come with a drink. Let's see the models for these tables:

// Meal model a record of meals table
type Meal struct {
    ID          uint
    Name        string
    Description string
    DrinkID     uint
    Drink       Drink
}

type Meals []Meal

// Drink model a record of drinks table
type Drink struct {
    ID          uint
    Name        string
    Description string
}

type Drinks []Drink

Now let's see the N+1 in action, here we have some methods to query data from the database (you can find the full code on my repo)

type Waitress struct {
    db *sql.DB
}

// getMeals gets 20 meals from the db
func (b Waitress) getMeals() (Meals, error) {
    // this method has the logic to query from the db, if you want to see
    // the actual code, you can go the repo

    // SQL used in this method
    // `SELECT id, name, description, drink_id FROM meals LIMIT 20`

    return meals, nil
}

// getDrinkByID gets a drink from the db by an id
func (b Waitress) getDrinkByID(ID uint) (Drink, error) {
    // this method has the logic to query from the db, if you want to see
    // the actual code, you can go the repo

    // SQL used in this method
  // `SELECT id, name, description FROM drinks WHERE id = $1`

    return Drink{}, nil
}

And finally the method with the N+1 query problem

func (b Waitress) ListMenu() (Meals, error) {
    // the `1` of our `N+1`
    meals, err := b.getMeals()
    if err != nil {
        return nil, err
    }

    // here is our waitress going back and forth 
    // from floor 1 to 10, this will be the `N` of our `N+1`
    for i, meal := range meals {
        drink, err := b.getDrinkByID(meal.DrinkID)
        if err != nil {
            return nil, err
        }

        meals[i].Drink = drink
    }

    return meals, nil
}

Yes, this simple logic can burn your database, because you are going back and forth to add the drinks to each meal, which is not efficient.

The more records you have to query or the most users you have, the most this N+1 problem we’ll affect your application because the time complexity is O(N)/Linear time.

ℹ️ Here I gave you an example in the backend, but this problem can also be found in your frontend, where instead of calling your database directly, you’ll be calling an endpoint of your backend which at the same time would call the database.

Solution

Now let's see two solutions to our problem.

1. Join the authors in the SQL query

   This maybe the easier solution, in here you’ll just have to write a query like the following:
   

   SELECT m.id, m.name, m.description, d.name, d.description
   FROM meals AS m
       INNER JOIN drinks AS d ON d.id = m.drink_id
   

   With this query, our code will look the following
   

   type Solution1 struct {
       db *sql.DB
   }
   
   func (s Solution1) ListMenu() (Meals, error) {
       // this method will execute the previous SQL
       // SELECT
       // m.id,
       //  m.name,
       //  m.description,
       //  d.name,
       //  d.description
       // FROM meals AS m
       // INNER JOIN drinks AS d ON d.id = m.drink_id
   
       // if you want to see the full code, refer to my repo
   
       return meals, nil
   }
   

   With this query, now we only have you query our database once and that’s it.

2. Get the meals and then join the drinks with your programming language

   No, we won’t do the same as the example where we saw the N+1 problem, here, instead of querying the meals and then querying the drinks one by one, we’ll just do two queries to our database, let see how:
   

   
   func (s Solution2) ListMenu() (Meals, error) {
       meals, err := s.getMeals()
       if err != nil {
           return nil, err
       }
   
       // this is our waitress bringing everything in a bussing cart 
       drinks, err := s.getDrinksByIDsIn(meals.GetUniqueDrinkIDs())
       if err != nil {
           return nil, err
       }
   
       meals.JoinDrinks(drinks)
   
       return meals, nil
   }
   
   func (s Solution2) getMeals() (Meals, error) {
       // this method has the logic to query from the db, if you want to see
       // what's in these methods, you can go the repo
   
       // sql used in this method
       // `SELECT id, name, description, drink_id FROM meals LIMIT 20`
   
       return Meals{}, nil
   }
   
   func (s Solution2) getDrinksByIDsIn(ids []uint) (Drinks, error) {
       // this method has the logic to query from the db, if you want to see
       // what's in these methods, you can go the repo
   
       // sql used in this method
       // `SELECT id, name, description FROM drinks WHERE id IN (1, 2, 3)`
   
       return Drinks{}, nil
   }
   

   As you can see, we only have two queries to our database: s.getMeals() and s.getDrinksByIDsIn, and if you read the ListMenu method, you’ve noticed that we introduced to more methods, let see what they do and why we need them:
   

   // to query the drinks, we need to know their IDs, for that
   // we add this method to our `Meals` slice
   func (m Meals) GetUniqueDrinkIDs() []uint {
       var ids []uint
       drinks := make(map[uint]struct{}, 0)
   
       for _, v := range m {
           _, ok := drinks[v.DrinkID]
           if ok {
               continue
           }
   
           drinks[v.DrinkID] = struct{}{}
           ids = append(ids, v.DrinkID)
       }
   
       return ids
   }
   
   // once we've query our meals and drinks, we proceed to join the 
   // corresponding drink to our meals
   func (m Meals) JoinDrinks(drink Drinks) {
       for i, meal := range m {
           m[i].Drink = drink.GetByID(meal.DrinkID)
       }
   }
   
   // this methos is needed by the `JoinDrinks()` method
   // so it can find the drink for a meal 
   func (d Drinks) GetByID(ID uint) Drink {
       for _, drink := range d {
           if drink.ID == ID {
               return drink
           }
       }
   
       return Drink{}
   }
   

Now, you can see we don't query the database for every drink, instead, in one query we get all the meals, and in the other we query the drinks and then join them to the corresponding meals.

When to use one solution or the other?

Well, in this app, every meal includes just one drink, but what happened if a meal includes more than one drink?, in that scenario, the first solution can't help us, because the SQL query is going to repeat the record for every drink in a meal, so what we want to do, is use the second solution, when we first query the meals, and then get the drinks to join them to the corresponding meals

Personal experience

At work, we have a microservice that is in charge to cache a lot of data about the products we have, twice a day or at demand, and it used to take about 1 minute to cache all the data because of this problem, after we remove the N+1s, it went from 1 minute to 2 seconds!.

Conclusion

Don’t overestimate a simple logic like the N+1’s, you can fall into this problem easily, but also you can fix it easily, but if you don’t do it in time, your application performance will let you know overtime.

Something I didn’t mention is, the N+1 in ORMs like Gorm because I don’t have experience with these, but I will recommend if you are using ORMs, that you dig into the underlying code to see if you have this problem.

⚠️ The way I structure the code in this article is not meant to be a guide on how you should structure your code, it is made as simple as possible to focus on the problem and solution code.

Homework

If you are working on a project, or you already have projects in production, go check them out to get rid of any N+1 you find.

References

1. N+1 in Laravel: Here you can see the N+1 in Eloquent (an ORM for Laravel)
2. Figma Jam and IconDuck: To make the illustrations
3. Article repository: You can find the examples in my GitHub