Objection + Knex = Painless PostgreSQL in your Node App

It's no secret that I'm a total PostgreSQL fangirl -- I rarely see a use case for using a different database, especially with the support for array and JSON fields. I also love Node and Express for simple APIs (without auth). In the past, SQL support within Node and Express hasn't been perfect. I've been hearing great things about Objection, so I decided to try it out!

Objection, which is built on top of Knex, uses the new ES7 class features to build a nice ORM query language for Node. ORMs allow you to use whatever programming language you are using for your app to query a database rather than querying in the natie language of the database (here we will use JavaScript to interact with our database instead of SQL). Since Objection is still really new, I will be walking through all of my code step by step.

The Learning Process

For this project, I relied pretty much exclusively on the documentation. The Knex documentation was great, and there were examples on the Objection GitHub that were very helpful as well. Since I make so many Express apps, given that I teach Express pretty extensively to my students, I felt pretty comfortable continuing with the project after skimming these resources.

The Final Project

I've been having trouble coming up with app ideas for this blog! So, I built an app idea app! The models were relatively simple: ideas and comments, but they still demonstrate one of the biggest use cases for Objection: relations between data. The ideas will be the "parents" with "child" comments attached to them. Essentially, users will be able to comment on various app ideas.

Knex Initialization

First, I initialized Knex, which will facilitate our database connection using pg, our migrations, and our seeds. After setting up my typical Express API boilerplate in my index file and installing the requirements in my package.json, I ran knex init in the root of my project. This created a knexfile.js that contains a boilerplate with example connections to databases. I decided to remove the production, development, and staging options in favor of just specifying a database connection string in my .env file. The knexfile ended up looking like:

require('dotenv').config()

const pg = require('pg')
pg.defaults.ssl = true

module.exports = {
  client: 'pg',
  connection: process.env.DATABASE_URL
}

The ssl configuration is only necessary if you are using a database on Heroku or another provider that requires an SSL connection. dotenv allows us to retrieve environmental variables from a .env file! That variable is a standard PostgreSQL connection string:

DATABASE_URL=postgres://username:password@host:port/db_name

I created the database on my computer using psql, I created the production database using a Heroku add-on.

Migrations

Migrations are changes to a database's schema specified within your ORM, so we will be defining the tables and columns of our database straight in JavaScript rather than using SQL.

From there, I generated my migrations:

$ knex migrate:make create_ideas
$ knex migrate:make create_comments

Each migrate command created its own separate file in the migrations folder. Knex also puts a timestamp on each so that the migration name is unique and is run in order, for example: migrations/20180218215453_create_ideas.js. I created two separate migrations to keep things organized, and because I created the comments after the ideas. These could be combined, though.

The migration is generated with:

exports.up = function (knex, Promise) {
}

exports.down = function (knex, Promise) {
}

The migration itself goes within the body of the exports.up function and then whatever the opposite of that migration does goes within exports.down. The exports.down allows us to undo migrations that we no longer want. For the create_ideas migration, I added the following:

exports.up = function (knex, Promise) {
  return Promise.all([
    knex.schema.createTable('ideas', table => {
      table.increments('id').primary()
      table.string('idea')
      table.string('creator')
    })
  ])
}

exports.down = function (knex, Promise) {
  return Promise.all([
    knex.schema.dropTable('ideas')
  ])
}

Knex migration functions should "always return a promise" according to its documentation. We can use Promise.all() in order to return an array of promises to resolve. Even though each function only has one action in this case, I could have added more actions separated by ,'s. The exports.up contains the table creation logic for the ideas table, including a primary key that is auto-incremented table.increments('id').primary(). It also has two other string columns called idea and creator. To undo the migration, we would drop the ideas table, as specified in the exports.down function.

The second migration to create the comments file is similar:

exports.up = function (knex, Promise) {
  return Promise.all([
    knex.schema.createTable('comments', table => {
      table.increments('id').primary()
      table.string('comment')
      table.string('creator')
      table.integer('ideas_id').references('ideas.id')
    })
  ])
}

exports.down = function (knex, Promise) {
  return Promise.all([
    knex.schema.dropTable('comments')
  ])
}

This migration looks very similar to the ideas one, the only difference is the foreign key: table.integer('ideas_id').references('ideas.id'). There are many ways to do this specified in the documentation; however, the Objection documentation does it this way so I did as well. Knex enforced the column name ideas_id rather than idea_id which was unsemantic. I am sure there is a way around that naming mandate; however, I didn't put much effort into looking it up!

Finally, I ran the migrations using the command:

$ knex migrate:latest

Even though the command implies it runs only the latest migration, it instead runs all migrations that haven't been run yet.

Database Seeding

Knex also has some built-in functionality to help us seed, or add initial test data, to our database.

$ knex seed:make ideas

The above command created a seeds directory with an ideas.js file within it. That file also had the following code in it:

exports.seed = function (knex, Promise) {
}

I added the following:

exports.seed = function (knex, Promise) {
  return knex('ideas').del().then(() => {
    return knex('ideas').insert([
        {creator: 'Ali', idea: 'A To Do List app!'},
        {creator: 'Ali', idea: 'A Blog!'},
        {creator: 'Ali', idea: 'A calculator'}
    ])
  })
}

This cleared the ideas table, so there wasn't any data in the table, and then it inserted three records into the database. It used the JSON keys and values to create those rows. I only seeded the ideas table, but you could definitely seed the comments table as well!

I then ran the following command to update the database:

$ knex seed:run

Models

Up until this point, we have been using Knex to interact with our database. Now, we are going to create some models using Objection in order to deal with the relationships between our database tables and to make our querying more explicit! I created a models folder with a schema.js file within it. You could structure this pretty much anyway -- one good way would be to have each model in a different file. I kept everything together, though, for demonstration's sake!

First, let's take care of some administrative things at the top:

const Knex = require('knex')
const connection = require('../knexfile')
const { Model } = require('objection')

const knexConnection = Knex(connection)

Model.knex(knexConnection)

These lines of code connect us to the database using our knexfile from earlier. We are also attaching Objection to our database connection.

Now, let's create our model for our Comment data. The models will allow us to interact cleanly with the data we are retrieving from our database.

class Comment extends Model {
  static get tableName () {
    return 'comments'
  }

  static get relationMappings () {
    return {
      idea: {
        relation: Model.BelongsToOneRelation,
        modelClass: Idea,
        join: {
          from: 'comments.ideas_id',
          to: 'ideas.id'
        }
      }
    }
  }
}

Let's break this down. The static getter method tableName returns the name comments: the name of the database table we want our Comment class to model! We also have a second static getter method that defines the Comment model's relationships to other models. In this case, the key of the outside object idea is how we will refer to the parent class. The relation key within the child object has the value Model.BelongsToOneRelation which says that each comment is going to have one parent idea. The modelClass says that the idea is coming from the Idea model and then the join specifies the database table and column names to perform a SQL join on, in this case, the ideas_id column in the comments table to the id column in the ideas table. static and get were added in ES6!

The Idea class looks almost identical, though the relationships are inverted!

class Idea extends Model {
  static get tableName () {
    return 'ideas'
  }

  static get relationMappings () {
    return {
      comments: {
        relation: Model.HasManyRelation,
        modelClass: Comment,
        join: {
          from: 'ideas.id',
          to: 'comments.ideas_id'
        }
      }
    }
  }
}

module.exports = { Idea, Comment }

In this case, our relationship is Model.HasManyRelation since one idea can have multiple comments! I also exported the models so they could be used within our other files.

Querying

The final file I worked with was controllers/ideas.js. I usually separate all my "controller" functions -- the routing functions that decide what each route renders -- into a file or files if there are lots of them! This week, I built an API that I will build a front-end for in the future.

First, some imports:

const express = require('express')

const { Idea, Comment } = require('../models/schema')

const router = express.Router()

Let's walk through the first method, a get request that returns all of the ideas:

router.get('/', async (req, res) => {
  const ideas = await Idea.query()
  res.json(ideas)
})

In the above example, we are making the arrow function callback that handles the request and response asynchronous using async, then we can "pause" the body of the function until the promise from our Idea.query() resolves. That query will return a JavaScript object with all of the items in our ideas table using our res.json(ideas) method. If we navigate to localhost:3000/ideas locally or https://application-ideas.herokuapp.com/ideas in production we see:

[
    {
        "id": 1,
        "idea": "A To Do List app!",
        "creator": "Ali"
    },
    {
        "id": 2,
        "idea": "A Blog!",
        "creator": "Ali"
    },
    {
        "id": 3,
        "idea": "A calculator",
        "creator": "Ali"
    }
]

Note: The Objection documentation uses async and await to handle promises in JavaScript; however, we could rewrite the above function to look like the following and that would work equally as well!

router.get('/', (req, res) => {
  Idea.query().then(ideas => {
    res.json(ideas)
  })
})

Instead of going through the other routes in paragraph form, I am going to put the annotated code below:

router.get('/:id', async (req, res) => {
  // gets one idea, found by id.
  //Also fetches the related comments using the .eager method
  const idea = await Idea.query().findById(req.params.id).eager('comments')
  res.json(idea)
})

router.post('/', async (req, res) => {
  // creates a new idea from the request body
  // only allows the idea and creator fields for safety
  const newIdea = req.body

  const idea = await Idea.query()
                         .allowInsert('[idea, creator]')
                         .insert(newIdea)

  res.send(idea)
})

router.post('/:id/comments', async (req, res) => {
  // creates a new comment that is a child of an idea, again sanitizes fields.
  const idea = await Idea.query().findById(req.params.id)

  await idea.$relatedQuery('comments')
            .allowInsert('[comment, creator]')
            .insert(req.body)

  res.send(idea)
})

router.delete('/:id', async (req, res) => {
  // deletes an idea
  await Idea.query().deleteById(req.params.id)

  res.redirect('/ideas')
})

router.delete('/:id/comments/:commentId', async (req, res) => {
  // deletes a comment
  await Comment.query().deleteById(req.params.commentId)

  res.redirect(`/ideas/${req.params.id}`)
})

module.exports = router

There's a bunch more you can do with Objection, like raw queries, interaction with JSON fields, and validations.

Next Steps

I had a really fun time working with Objection and Knex! It is honestly very similar to working with Mongoose and MongoDB from a configuration standpoint, but it makes hierarchical and related data so much easier to work with! I would definitely keep using these libraries in the future with Express apps! Definitely a must-try if you use Node frequently!

Coming next week: a front-end for this API! Please leave a comment or tweet me with suggestions for a front-end tool to use for it!

Full Code
Deployed App
Objection Documentation
Knex Documentation

Part of my On Learning New Things Series

Comments

[Sami Koskimäki]

Great post, thanks for writing it! You have a small error in the usage of allowInsert method though. It can only be used to limit relations when inserting/upserting graphs using insertGraph and upsertGraph methods. It has no effect with the normal insert method.

[Daniel HB]

How does Knex compare to Sequelize?

[Dian Fay]

Like apples and oranges, basically. Sequelize, like its peers Objection and Bookshelf, is a full-fledged O/RM which gives you a well-defined, database-agnostic API at the cost of maintaining parallel data models, somewhat restricted flexibility (the stateful entity metaphor O/RMs encourage is a very limited view of what your database can actually do), lowest-common-denominator feature sets, and winding up writing SQL anyway when things get complicated, which they always do. As @rhymes mentioned below, I run a competing project with a different approach to data access, so take my opinion for what it's worth :)

Knex, meanwhile, is a query builder which generates SQL statements. It's lower-level and doesn't care about your data model; it's strictly a way to build SQL in JavaScript.

[Sami Koskimäki]

Based on this comment, I'm guessing you haven't actually used objection. The main design goal of objection is to NOT restrict flexibility or limit access to DB features.

[Dian Fay]

I have to admit I have not; the last time I had to search for a new data access framework for Node was a matter of days before you released 0.1.0 :) I did at least do the bare minimum of reading your documentation and looking at examples to see how it addressed the fundamental weaknesses of object-relational mapping before I mentioned it, though!

I consider models inherently limiting. To be completely fair to Objection, a minimally functional model is maybe half a dozen lines long, and patch takes care of one of the major restrictions of the active record pattern, but recapitulating your database architecture before you can use it necessarily adds a structural rigidity (if badly done, it can be outright brittle) and a certain inertia. This isn't something that can be laid at your or anyone else's doorstep; it's the price of working with record graphs in application code, as implicit in the object-relational mapper pattern. You can argue that it is or can be worth it, but that doesn't mean you don't have to pay it.

[Sami Koskimäki]

First of all, objection is built on knex and every possible operation you can start with it return a query builder that you can modify in any way. You are always 100% in control of what SQL gets executed. There's absolutely no restricting "common-denominator" API. With objection you can always do everything you can do with a query builder, but also a lot of other things. Objection's API is a superset of SQL instead of common-denominator.

Your other point was that models are limiting. Let's test that theory. Let's say you have this model in objection

class Person extends Model {
  static tableName = 'persons'
}

Then you have a bunch of queries all over your database

const { Person }  = require('./models/person')

...

Person.query().select('...').where('...')

How is that any different from using a query builder? Instead of mentioning the name of the table in the query, you use the model. The table name and the model class reference are just as easy to change.

Let's say the name of the table changes. All you need to do is to change the tableName property. What if the columns names change? You change the queries, just like you would with a query builder. What if you want to change the model name? Well, search and replace the model name. How about relations? Again, change the model. If you use a plain query builder, you need to update the foreign keys in million places instead.

Using objection's models adds no overhead on top of using a query builder whatsoever. If anything, it removes a great deal of it.

I really don't see what you mean here. You still seem to be talking about your experiences with other ORMs.

[krckyboy]

I have to say that Objection.js has by far the best docs that I've ever seen so far. Good job!

[Antonin J. (they/them)]

This is really cool! I actually use Postgres/Knex/Objection at work in production, and have for about a year and a half and it's an amazing tool. One of my favorite features of Objection is the use of the get and set in the class which you mentioned.

Combined with lodash, it's really easy to do something like:

res.send(_.pick(post, [
  ...post,
  computedProperty,
  computedProperty2,
  whateverElse
]);

I wanted to ask, why did you end up trying out Objection over, let's say, Bookshelf which was built by the same author as Knex and at around the same time as Knex? We migrated from Bookshelf to Objection so I think Objection is the right choice, just wondering though.

[Ali Spittel]

Ah awesome! I had a friend recommend Objection, so that's why I went with that one! I hadn't heard great things about Bookshelf.

[Lars Wächter]

Perfect timing! I like to add ORM to my node apps in the next days as well. Thanks

[rhymes]

Now that you tested an ORM library I guess you have to test @dmfay Massive.js based on the Data Mapper pattern as well :-D

Great post, as usual!

[Ali Spittel]

Ah! Will look into that!

[Sholto Maud]

Nice Post Ali,

Do you have a link to how best integrate "auth"?

[Ali Spittel]

Ah that would be awesome! I looked a little bit but didn't see anything. I think you could use Passport, though (kinda like this: code.tutsplus.com/tutorials/using-...)

[Belhassen Chelbi]

loved it

[Gleb Bahmutov]

I highly recommend using "knex-migrate" github.com/sheerun/knex-migrate#re... to do Knex migrations - a very nice tool with better commands for running / rolling / seeing migrations.

[Ali Spittel]

Oh cool! Didn't know this existed!

[Manuel Romero]

Great post, Ali!

[Arlei F. Farnetani Junior]

Excellent post! Thanks a lot!