The Flask Mega-Tutorial Part VIII: Followers (2018)

Posted by
on under

(Great news! There is a new version of this tutorial!)

This is the eighth installment of the Flask Mega-Tutorial series, in which I'm going to tell you how to implement a "followers" feature similar to that of Twitter and other social networks.

For your reference, below is a list of the articles in this series.

In this chapter I am going to work on the application's database some more. I want users of the application to be able to easily choose which other users they want to follow. So I'm going to be expanding the database so that it can keep track of who is following who, which is trickier than you may think.

The GitHub links for this chapter are: Browse, Zip, Diff.

Database Relationships Revisited

I said above that I want to maintain a list of "followed" and "follower" users for each user. Unfortunately, a relational database does not have a list type that I can use for these lists, all there is are tables with records and relationships between these records.

The database has a table that represents users, so what's left is to come up with the proper relationship type that can model the follower/followed link. This is a good time to review the basic database relationship types:


I have already used a one-to-many relationship in Chapter 4. Here is the diagram for this relationship:

One-to-many Relationship

The two entities linked by this relationship are users and posts. I say that a user has many posts, and a post has one user (or author). The relationship is represented in the database with the use of a foreign key on the "many" side. In the relationship above, the foreign key is the user_id field added to the posts table. This field links each post to the record of its author in the user table.

It is pretty clear that the user_id field provides direct access to the author of a given post, but what about the reverse direction? For the relationship to be useful I should be able to get the list of posts written by a given user. The user_id field in the posts table is also sufficient to answer this question, as databases have indexes that allow for efficient queries such us "retrieve all posts that have a user_id of X".


A many-to-many relationship is a bit more complex. As an example, consider a database that has students and teachers. I can say that a student has many teachers, and a teacher has many students. It's like two overlapped one-to-many relationships from both ends.

For a relationship of this type I should be able to query the database and obtain the list of teachers that teach a given student, and the list of students in a teacher's class. This is actually non-trivial to represent in a relational database, as it cannot be done by adding foreign keys to the existing tables.

The representation of a many-to-many relationship requires the use of an auxiliary table called an association table. Here is how the database would look for the students and teachers example:


While it may not seem obvious at first, the association table with its two foreign keys is able to efficiently answer all the queries about the relationship.

Many-to-One and One-to-One

A many-to-one is similar to a one-to-many relationship. The difference is that this relationship is looked at from the "many" side.

A one-to-one relationship is a special case of a one-to-many. The representation is similar, but a constraint is added to the database to prevent the "many" side to have more than one link. While there are cases in which this type of relationship is useful, it isn't as common as the other types.

Representing Followers

Looking at the summary of all the relationship types, it is easy to determine that the proper data model to track followers is the many-to-many relationship, because a user follows many users, and a user has many followers. But there is a twist. In the students and teachers example I had two entities that were related through the many-to-many relationship. But in the case of followers, I have users following other users, so there is just users. So what is the second entity of the many-to-many relationship?

The second entity of the relationship is also the users. A relationship in which instances of a class are linked to other instances of the same class is called a self-referential relationship, and that is exactly what I have here.

Here is a diagram of the self-referential many-to-many relationship that keeps track of followers:


The followers table is the association table of the relationship. The foreign keys in this table are both pointing at entries in the user table, since it is linking users to users. Each record in this table represents one link between a follower user and a followed user. Like the students and teachers example, a setup like this one allows the database to answer all the questions about followed and follower users that I will ever need. Pretty neat.

Database Model Representation

Let's add followers to the database first. Here is the followers association table:

app/ Followers association table

followers = db.Table('followers',
    db.Column('follower_id', db.Integer, db.ForeignKey('')),
    db.Column('followed_id', db.Integer, db.ForeignKey(''))

This is a direct translation of the association table from my diagram above. Note that I am not declaring this table as a model, like I did for the users and posts tables. Since this is an auxiliary table that has no data other than the foreign keys, I created it without an associated model class.

Now I can declare the many-to-many relationship in the users table:

app/ Many-to-many followers relationship

class User(UserMixin, db.Model):
    # ...
    followed = db.relationship(
        'User', secondary=followers,
        primaryjoin=(followers.c.follower_id == id),
        secondaryjoin=(followers.c.followed_id == id),
        backref=db.backref('followers', lazy='dynamic'), lazy='dynamic')

The setup of the relationship is non-trivial. Like I did for the posts one-to-many relationship, I'm using the db.relationship function to define the relationship in the model class. This relationship links User instances to other User instances, so as a convention let's say that for a pair of users linked by this relationship, the left side user is following the right side user. I'm defining the relationship as seen from the left side user with the name followed, because when I query this relationship from the left side I will get the list of followed users (i.e those on the right side). Let's examine all the arguments to the db.relationship() call one by one:

  • 'User' is the right side entity of the relationship (the left side entity is the parent class). Since this is a self-referential relationship, I have to use the same class on both sides.
  • secondary configures the association table that is used for this relationship, which I defined right above this class.
  • primaryjoin indicates the condition that links the left side entity (the follower user) with the association table. The join condition for the left side of the relationship is the user ID matching the follower_id field of the association table. The value of this argument is followers.c.follower_id, which qreferences the follower_id column of the association table.
  • secondaryjoin indicates the condition that links the right side entity (the followed user) with the association table. This condition is similar to the one for primaryjoin, with the only difference that now I'm using followed_id, which is the other foreign key in the association table.
  • backref defines how this relationship will be accessed from the right side entity. From the left side, the relationship is named followed, so from the right side I am going to use the name followers to represent all the left side users that are linked to the target user in the right side. The additional lazy argument indicates the execution mode for this query. A mode of dynamic sets up the query to not run until specifically requested, which is also how I set up the posts one-to-many relationship.
  • lazy is similar to the parameter of the same name in the backref, but this one applies to the left side query instead of the right side.

Don't worry if this is hard to understand. I will show you how to work with these queries in a moment, and then everything will become clearer.

The changes to the database need to be recorded in a new database migration:

(venv) $ flask db migrate -m "followers"
INFO  [alembic.runtime.migration] Context impl SQLiteImpl.
INFO  [alembic.runtime.migration] Will assume non-transactional DDL.
INFO  [] Detected added table 'followers'
  Generating /home/miguel/microblog/migrations/versions/ ... done

(venv) $ flask db upgrade
INFO  [alembic.runtime.migration] Context impl SQLiteImpl.
INFO  [alembic.runtime.migration] Will assume non-transactional DDL.
INFO  [alembic.runtime.migration] Running upgrade 37f06a334dbf -> ae346256b650, followers

Adding and Removing "follows"

Thanks to the SQLAlchemy ORM, a user following another user can be recorded in the database working with the followed relationship as if it was a list. For example, if I had two users stored in user1 and user2 variables, I can make the first follow the second with this simple statement:


To stop following the user, then I could do:


Even though adding and removing followers is fairly easy, I want to promote reusability in my code, so I'm not going to sprinkle "appends" and "removes" through the code. Instead, I'm going to implement the "follow" and "unfollow" functionality as methods in the User model. It is always best to move the application logic away from view functions and into models or other auxiliary classes or modules, because as you will see later in this chapter, that makes unit testing much easier.

Below are the changes in the user model to add and remove relationships:

app/ Add and remove followers

class User(UserMixin, db.Model):

    def follow(self, user):
        if not self.is_following(user):

    def unfollow(self, user):
        if self.is_following(user):

    def is_following(self, user):
        return self.followed.filter(
            followers.c.followed_id == > 0

The follow() and unfollow() methods use the append() and remove() methods of the relationship object as I have shown above, but before they touch the relationship they use the is_following() supporting method to make sure the requested action makes sense. For example, if I ask user1 to follow user2, but it turns out that this following relationship already exists in the database, I do not want to add a duplicate. The same logic can be applied to unfollowing.

The is_following() method issues a query on the followed relationship to check if a link between two users already exists. You have seen me use the filter_by() method of the SQLAlchemy query object before, for example to find a user given its username. The filter() method that I'm using here is similar, but lower level, as it can include arbitrary filtering conditions, unlike filter_by() which can only check for equality to a constant value. The condition that I'm using in is_following() looks for items in the association table that have the left side foreign key set to the self user, and the right side set to the user argument. The query is terminated with a count() method, which returns the number of results. The result of this query is going to be 0 or 1, so checking for the count being 1 or greater than 0 is actually equivalent. Other query terminators you have seen me use in the past are all() and first().

Obtaining the Posts from Followed Users

Support for followers in the database is almost complete, but I'm actually missing one important feature. In the index page of the application I'm going to show blog posts written by all the people that are followed by the logged in user, so I need to come up with a database query that returns these posts.

The most obvious solution is to run a query that returns the list of followed users, which as you already know, it would be user.followed.all(). Then for each of these returned users I can run a query to get the posts. Once I have all the posts I can merge them into a single list and sort them by date. Sounds good? Well, not really.

This approach has a couple of problems. What happens if a user is following a thousand people? I would need to execute a thousand database queries just to collect all the posts. And then I will need to merge and sort the thousand lists in memory. As a secondary problem, consider that the application's home page will eventually have pagination implemented, so it will not display all the available posts but just the first few, with a link to get more if desired. If I'm going to display posts sorted by their date, how can I know which posts are the most recent of all followed users combined, unless I get all the posts and sort them first? This is actually an awful solution that does not scale well.

There is really no way to avoid this merging and sorting of blog posts, but doing it in the application results in a very inefficient process. This kind of work is what relational databases excel at. The database has indexes that allow it to perform the queries and the sorting in a much more efficient way that I can possibly do from my side. So what I really want is to come up with a single database query that defines the information that I want to get, and then let the database figure out how to extract that information in the most efficient way.

Below you can see this query:

app/ Followed posts query

class User(UserMixin, db.Model):
    def followed_posts(self):
        return Post.query.join(
            followers, (followers.c.followed_id == Post.user_id)).filter(
                followers.c.follower_id ==

This is by far the most complex query I have used on this application. I'm going to try to decipher this query one piece at a time. If you look at the structure of this query, you are going to notice that there are three main sections designed by the join(), filter() and order_by() methods of the SQLAlchemy query object:



To understand what a join operation does, let's look at an example. Let's assume that I have a User table with the following contents:

id username
1 john
2 susan
3 mary
4 david

To keep things simple I am not showing all the fields in the user model, just the ones that are important for this query.

Let's say that the followers association table says that user john is following users susan and david, user susan is following mary and user mary is following david. The data that represents the above is this:

follower_id followed_id
1 2
1 4
2 3
3 4

Finally, the posts table contains one post from each user:

id text user_id
1 post from susan 2
2 post from mary 3
3 post from david 4
4 post from john 1

This table also omits some fields that are not part of this discussion.

Here is the join() call that I defined for this query once again:

Post.query.join(followers, (followers.c.followed_id == Post.user_id))

I'm invoking the join operation on the posts table. The first argument is the followers association table, and the second argument is the join condition. What I'm saying with this call is that I want the database to create a temporary table that combines data from posts and followers tables. The data is going to be merged according to the condition that I passed as argument.

The condition that I used says that the followed_id field of the followers table must be equal to the user_id of the posts table. To perform this merge, the database will take each record from the posts table (the left side of the join) and append any records from the followers table (the right side of the join) that match the condition. If multiple records in followers match the condition, then the post entry will be repeated for each. If for a given post there is no match in followers, then that post record is not part of the join.

With the example data I defined above, the result of the join operation is:

id text user_id follower_id followed_id
1 post from susan 2 1 2
2 post from mary 3 2 3
3 post from david 4 1 4
3 post from david 4 3 4

Note how the user_id and followed_id columns are equal in all cases, as this was the join condition. The post from user john does not appear in the joined table because there are no entries in followers that have john as a followed user, or in other words, nobody is following john. And the post from david appears twice, because that user is followed by two different users.

It may not be immediately clear what do I gain by creating this join, but keep reading, as this is just one part of the bigger query.


The join operation gave me a list of all the posts that are followed by some user, which is a lot more data that I really want. I'm only interested in a subset of this list, the posts followed by a single user, so I need trim all the entries I don't need, which I can do with a filter() call.

Here is the filter portion of the query:

filter(followers.c.follower_id ==

Since this query is in a method of class User, the expression refers to the user ID of the user I'm interested in. The filter() call selects the items in the joined table that have the follower_id column set to this user, which in other words means that I'm keeping only the entries that have this user as a follower.

Let's say the user I'm interested in is john, which has its id field set to 1. Here is how the joined table looks after the filtering:

id text user_id follower_id followed_id
1 post from susan 2 1 2
3 post from david 4 1 4

And these are exactly the posts that I wanted!

Remember that the query was issued on the Post class, so even though I ended up with a temporary table that was created by the database as part of this query, the result will be the posts that are included in this temporary table, without the extra columns added by the join operation.


The final step of the process is to sort the results. The part of the query that does that says:


Here I'm saying that I want the results sorted by the timestamp field of the post in descending order. With this ordering, the first result will be the most recent blog post.

Combining Own and Followed Posts

The query that I'm using in the followed_posts() function is extremely useful, but has one limitation. People expect to see their own posts included in their timeline of followed users, and the query as it is does not have that capability.

There are two possible ways to expand this query to include the user's own posts. The most straightforward way is to leave the query as it is, but make sure all users are following themselves. If you are your own follower, then the query as shown above will find your own posts along with those of all the people you follow. The disadvantage of this method is that it affects the stats regarding followers. All follower counts are going to be inflated by one, so they'll have to be adjusted before they are shown. The second way to do this is by create a second query that returns the user's own posts, and then use the "union" operator to combine the two queries into a single one.

After considering both options I decided to go with the second one. Below you can see the followed_posts() function after it has been expanded to include the user's posts through a union:

app/ Followed posts query with user's own posts.

    def followed_posts(self):
        followed = Post.query.join(
            followers, (followers.c.followed_id == Post.user_id)).filter(
                followers.c.follower_id ==
        own = Post.query.filter_by(
        return followed.union(own).order_by(Post.timestamp.desc())

Note how the followed and own queries are combined into one, before the sorting is applied.

Unit Testing the User Model

While I don't consider the followers implementation I have built a "complex" feature, I think it is also not trivial. My concern when I write non-trivial code, is to ensure that this code will continue to work in the future, as I make modifications on different parts of the application. The best way to ensure that code you have already written continues to work in the future is to create a suite of automated tests that you can re-run each time changes are made.

Python includes a very useful unittest package that makes it easy to write and execute unit tests. Let's write some unit tests for the existing methods in the User class in a module: User model unit tests.

import os
os.environ['DATABASE_URL'] = 'sqlite://'

from datetime import datetime, timedelta
import unittest
from app import app, db
from app.models import User, Post

class UserModelCase(unittest.TestCase):
    def setUp(self):
        self.app_context = app.app_context()

    def tearDown(self):

    def test_password_hashing(self):
        u = User(username='susan')

    def test_avatar(self):
        u = User(username='john', email='')
        self.assertEqual(u.avatar(128), (''

    def test_follow(self):
        u1 = User(username='john', email='')
        u2 = User(username='susan', email='')
        self.assertEqual(u1.followed.all(), [])
        self.assertEqual(u1.followers.all(), [])

        self.assertEqual(u1.followed.count(), 1)
        self.assertEqual(u1.followed.first().username, 'susan')
        self.assertEqual(u2.followers.count(), 1)
        self.assertEqual(u2.followers.first().username, 'john')

        self.assertEqual(u1.followed.count(), 0)
        self.assertEqual(u2.followers.count(), 0)

    def test_follow_posts(self):
        # create four users
        u1 = User(username='john', email='')
        u2 = User(username='susan', email='')
        u3 = User(username='mary', email='')
        u4 = User(username='david', email='')
        db.session.add_all([u1, u2, u3, u4])

        # create four posts
        now = datetime.utcnow()
        p1 = Post(body="post from john", author=u1,
                  timestamp=now + timedelta(seconds=1))
        p2 = Post(body="post from susan", author=u2,
                  timestamp=now + timedelta(seconds=4))
        p3 = Post(body="post from mary", author=u3,
                  timestamp=now + timedelta(seconds=3))
        p4 = Post(body="post from david", author=u4,
                  timestamp=now + timedelta(seconds=2))
        db.session.add_all([p1, p2, p3, p4])

        # setup the followers
        u1.follow(u2)  # john follows susan
        u1.follow(u4)  # john follows david
        u2.follow(u3)  # susan follows mary
        u3.follow(u4)  # mary follows david

        # check the followed posts of each user
        f1 = u1.followed_posts().all()
        f2 = u2.followed_posts().all()
        f3 = u3.followed_posts().all()
        f4 = u4.followed_posts().all()
        self.assertEqual(f1, [p2, p4, p1])
        self.assertEqual(f2, [p2, p3])
        self.assertEqual(f3, [p3, p4])
        self.assertEqual(f4, [p4])

if __name__ == '__main__':

I have added four tests that exercise the password hashing, user avatar and followers functionality in the user model. The setUp() and tearDown() methods are special methods that the unit testing framework executes before and after each test respectively.

I have implemented a little hack to prevent the unit tests from using the regular database that I use for development. By setting the DATABASE_URL environment variable to sqlite://, I change the application configuration to direct SQLAlchemy to use an in-memory SQLite database during the tests.

The setUp() method then creates an application context and pushes it. This ensures that the Flask application instance, along with its configuration data is accessible to Flask extensions. Don't worry if this does not make a lot of sense at this point, as this will be covered in more detail later.

The db.create_all() call creates all the database tables. This is a quick way to create a database from scratch that is useful for testing. For development and production use I have already shown you how to create database tables through database migrations.

You can run the entire test suite with the following command:

(venv) $ python
test_avatar (__main__.UserModelCase) ... ok
test_follow (__main__.UserModelCase) ... ok
test_follow_posts (__main__.UserModelCase) ... ok
test_password_hashing (__main__.UserModelCase) ... ok

Ran 4 tests in 0.494s


From now on, every time a change is made to the application, you can re-run the tests to make sure the features that are being tested have not been affected. Also, each time another feature is added to the application, a unit test should be written for it.

Integrating Followers with the Application

The support of followers in the database and models is now complete, but I don't have any of this functionality incorporated into the application, so I'm going to add that now.

Because the follow and unfollow actions introduce changes in the application, I'm going to implement them as POST requests, which are triggered from the web browser as a result of submitting a web form. It would be easier to implement these routes as GET requests, but then they could be exploited in CSRF attacks. Because GET requests are harder to protect against CSRF, they should only be used on actions that do not introduce state changes. Implementing these as a result of a form submission is better because then a CSRF token can be added to the form.

But how can a follow or unfollow action be triggered from a web form when the only thing the user needs to do is click on "Follow" or "Unfollow", without submitting any data? To make this work, the form is going to be empty. The only elements in the form are going to be the CSRF token, which is implemented as a hidden field and added automatically by Flask-WTF, and a submit button, which is going to be what the user needs to click to trigger the action. Since the two actions are almost identical I'm going to use the same form for both. I'm going to call this form EmptyForm.

app/ Empty form for following and unfollowing.

class EmptyForm(FlaskForm):
    submit = SubmitField('Submit')

Let's add two new routes in the application to follow and unfollow a user:

app/ Follow and unfollow routes.

from app.forms import EmptyForm

# ...

@app.route('/follow/<username>', methods=['POST'])
def follow(username):
    form = EmptyForm()
    if form.validate_on_submit():
        user = User.query.filter_by(username=username).first()
        if user is None:
            flash('User {} not found.'.format(username))
            return redirect(url_for('index'))
        if user == current_user:
            flash('You cannot follow yourself!')
            return redirect(url_for('user', username=username))
        flash('You are following {}!'.format(username))
        return redirect(url_for('user', username=username))
        return redirect(url_for('index'))

@app.route('/unfollow/<username>', methods=['POST'])
def unfollow(username):
    form = EmptyForm()
    if form.validate_on_submit():
        user = User.query.filter_by(username=username).first()
        if user is None:
            flash('User {} not found.'.format(username))
            return redirect(url_for('index'))
        if user == current_user:
            flash('You cannot unfollow yourself!')
            return redirect(url_for('user', username=username))
        flash('You are not following {}.'.format(username))
        return redirect(url_for('user', username=username))
        return redirect(url_for('index'))

The form handling in these routes is simpler, because we only have to implement the submission part. Unlike other forms such as the login and edit profile forms, these two forms do not have their own pages, the forms will be rendered by the user() route and will appear in the user's profile page. The only reason why the validate_on_submit() call can fail is if the CSRF token is missing or invalid, so in that case I just redirect the application back to the home page.

If the form validation passes, I do some error checking before actually carrying out the follow or unfollow action. This is to prevent unexpected issues, and to try to provide a useful message to the user when a problem has occurred.

To render the follow or unfollow button, I need to instantiate an EmptyForm object and pass it to the user.html template. Because these two actions are mutually exclusive, I can pass a single instance of this generic form to the template:

app/ Follow and unfollow routes.

def user(username):
    # ...
    form = EmptyForm()
    return render_template('user.html', user=user, posts=posts, form=form)

I can now add the follow or unfollow forms in the profile page of each user:

app/templates/user.html: Follow and unfollow links in user profile page.

        <h1>User: {{ user.username }}</h1>
        {% if user.about_me %}<p>{{ user.about_me }}</p>{% endif %}
        {% if user.last_seen %}<p>Last seen on: {{ user.last_seen }}</p>{% endif %}
        <p>{{ user.followers.count() }} followers, {{ user.followed.count() }} following.</p>
        {% if user == current_user %}
        <p><a href="{{ url_for('edit_profile') }}">Edit your profile</a></p>
        {% elif not current_user.is_following(user) %}
            <form action="{{ url_for('follow', username=user.username) }}" method="post">
                {{ form.hidden_tag() }}
                {{ form.submit(value='Follow') }}
        {% else %}
            <form action="{{ url_for('unfollow', username=user.username) }}" method="post">
                {{ form.hidden_tag() }}
                {{ form.submit(value='Unfollow') }}
        {% endif %}

The changes to the user profile template add a line below the last seen timestamp that shows how many followers and followed users this user has. And the line that has the "Edit" link when you are viewing your own profile now can have one of three possible links:

  • If the user is viewing his or her own profile, the "Edit" link shows as before.
  • If the user is viewing a user that is not currently followed, the "Follow" form shows.
  • If the user is viewing a user that is currently followed, the "Unfollow" form shows.

To reuse the EmptyForm() instance for both the follow and unfollow forms, I pass a value argument when rendering the submit button. In a submit button, the value attribute defines the label, so with this trick I can change the text in the submit button depending on the action that I need to present to the user.

At this point you can run the application, create a few users and play with following and unfollowing users. The only thing you need to remember is to type the profile page URL of the user you want to follow or unfollow, since there is currently no way to see a list of users. For example, if you want to follow a user with the susan username, you will need to type http://localhost:5000/user/susan in the browser's address bar to access the profile page for that user. Make sure you check how the followed and follower user counts change as you issue follows or unfollows.

I should be showing the list of followed posts in the index page of the application, but I don't have all the pieces in place to do that yet, since users cannot write blog posts yet. So I'm going to delay this change until that functionality is in place.

Become a Patron!

Hello, and thank you for visiting my blog! If you enjoyed this article, please consider supporting my work on this blog on Patreon!

  • #1 Ben Kozuch said

    Thank you so much! Best programming guide on the web!

  • #2 sunny said

    Hi Miguel,
    at first i want to say thank you for this tutorial. I have issue when i run, testcase "test_follow_posts" is failing with error "TypeError: 'author' is an invalid keyword argument for Post"

  • #3 Miguel Grinberg said

    @sunny: my guess is that you did not define the relationship from posts to users in the User model. That has a backref argument that defines "author".

  • #4 Rustam said

    I would like to suggest to add some spacing between colons in section where you start explanation of Join....idtext idusername looks a little confusing.....

  • #5 Miguel Grinberg said

    @Rustam: I think your browser may be caching an old version of my CSS. I have fixed the tables before publishing this article.

  • #6 Vitalii said

    Hi Miguel,
    Thanks for your posts. I have some question. It seems to me that follow and unfollow routes can be joined to meet DRY concept. We can just pass some kind of action type (follow/unfollow) as an argument from URL.Thoughts?

  • #7 Miguel Grinberg said

    @Vitalii: Yeah, that is a possibility. I decided to keep them separate to avoid having to add conditionals for the status and error messages, but if you prefer that and have it all in one function it's fine.

  • #8 McFate said

    Michael, this series is a great update on the original. One minor point that has me a little bit baffled here: in the expressions involving the followers table, such as "followers.c.followed_id", where does the "c" come from...? Why not "followers.followed_id"?

  • #9 Miguel Grinberg said

    @McFate: The "c" is an attribute of SQLAlchemy tables that are not defined as models. For these tables, the table columns are all exposed as sub-attributes of this "c" attribute.

  • #10 James said

    I keep getting the following when trying to update the database to add followers/followed. I tried copy/pasting the code exactly as listed and it still won't register a change to the DB. Anything obvious I could be missing?

    INFO [alembic.runtime.migration] Context impl SQLiteImpl.
    INFO [alembic.runtime.migration] Will assume non-transactional DDL.
    INFO [alembic.env] No changes in schema detected.

  • #11 Miguel Grinberg said

    James: The first two lines are informational and can be ignored. The third line says that there are no changes to your models, when compared against the current state of your database tables. So you either did not change the models to add followers, or if you did, those changes somehow made it to the database tables already.

  • #12 Andre said

    Hi Miguel,

    Thank you for the tutorials, those are really helpful. I am quite familar with Python, but struggle a bit with the database concepts. Could you point me to some good tutorials to start database modeling/design?

  • #13 Miguel Grinberg said

    @Andre: I have learned relational database ages ago, so I'm not super familiar with more recent tutorials. One option that you have is the book on SQLAlchemy by O'Reilly.

  • #14 Jacob K said


    Great tutorial so far. I have a problem though. I accidentally added the definition of the followers table to the User model instead of by itself and committed the schema changes to the db. I made the necessary adjustments to to get it in the right place, however, when I run the migrate command, it states there are no schema changes detected.

    What would be the best way to undo this?

  • #15 Miguel Grinberg said

    If you don't care about preserving the data in your database, then deleting the database file and bad migration script and running "flask db upgrade" should put you in the state where you can regenerate the followers migration.

  • #16 Ghouse said

    Hi miguel,

    After understanding followers concept.
    I observed that in user class you created a many to many relationship with auxilary table "followers". as below

    followed = db.relationship(
    'User', secondary=followers,
    primaryjoin=(followers.c.follower_id == id),
    secondaryjoin=(followers.c.followed_id == id),
    backref=db.backref('followers', lazy='dynamic'), lazy='dynamic')

    In above code we can write backref='followers' instead of backref=db.backref('followers',lazy='dynamic').the only difference i observed is as below.I'm just curious to know why you used backref=db.backref(...) is there any reason. Could you please guide me if missing anything here.

    Example 1: If we use backref='followers' we can access user followers as u1.followers
    Output: ipython console output
    In [23]: u1
    Out[24]: <user john>

    In [25]: u2
    Out[26]: <user sum>

    In [27]: u2.follow(u1)

    In [28]: db.session.commit()

    In [29]: u1.followers
    Out[29]: [<user sum>]

    Example 2: If we use backref=db.backref('followers',lazy='dynamic') we can access user followers as u1.followers.all()
    Output: ipython console output.
    In [4]: u1 = User(username='john', email='')

    In [5]: u2 = User(username='sum', email='')

    In [6]: db.session.add(u1)

    In [7]: db.session.add(u2)

    In [8]: db.session.commit()

    In [9]: u1.follow(u2)

    In [10]: u2.followers
    Out[11]: <sqlalchemy.orm.dynamic.AppenderBaseQuery at 0x2bb662c44a8>

    In [12]: u2.followers.all()
    Out[14]: [<user john>]

  • #17 Miguel Grinberg said

    @Ghouse: this is explained in the article. A regular backref provides the query results directly. A backref that is created with the lazy=dynamic argument returns a query instead of the final results. Getting the query is useful because you can expand the query before you execute it to obtain the results. For example, you can ask that the results from the relationship come sorted in a specific way.

  • #18 Ars said

    Question about obtaining the posts from followed users.
    For example, Django ORM has filtering with 'in' condition:

    Can we use same approach here, with sqlalchemy?
    kinda: session.query(Post).filter(,456))).all() # just an example
    Will this query hit db just once?

  • #19 Miguel Grinberg said

    @Ars: You can use the "in" operator with SQLAlchemy, if that's what you want, but your Django ORM based example does not issue one query, it issues two queries. So it is really not as optimal as the approach I present here based on joins.

  • #20 Andy said

    Hi Miguel,

    thank you so much for doing this tutorial, I have had a great time following it along, and a lot of things have become less obscure.

    One question regarding "User.is_following()". You mention that SQLAlchemy exposes the "followed" relationship as list. Why not check for list membership using the "in" operator, like so:

    def is_following(self, user):
    return user in self.followed

  • #21 Miguel Grinberg said

    @Andy: self.followed is a database query, not a list, so you definitely not want to do what you did, as that will run the query, get all the elements into memory, and then check if the user is in the list. It is a very inefficient way to do the check. My solution only retrieves the one element we are interested in, instead of the whole list.

  • #22 David Collins said

    Does the sequencing of .join(), .filter(), .order_by() matter? If so, how? Thanks!

  • #23 Miguel Grinberg said

    @David: I'm not sure this is true for every possible construct, but in general the order does not matter. You can check how the SQL looks like for a query with print(str(your_query_here)). So try that with changing the order of the query parts with your actual database to see if it has an effect.

  • #24 James said

    Hi Miguel. Thank you for your wonderful tutorial. Everything seemed to be working for me untill I come across this error in my code. The server starts well, but get into the following summarized error when I open the browser to access the app. For the last three days, I have stuggled with it, check code for code, up to this level but still cant exactly understand whey the error still occuring, yet the said table exist when I check manually. Here is the error:
    I have counter-checked every code you provided at this stage with mine but still gettionalng the error. sqlalchemy.exc.Operational Error: (sqlite3.OperationalError) no such table: user [SQL: 'SELECT AS user_id, AS user_namename, AS user_email, user.password_hash AS user_password_hash, user.about_me AS user_about_me, user.last_seen AS user_last_seen \nFROM user \nWHERE = ?][parameters(2,)] (Background on this error at

  • #25 Miguel Grinberg said

    @James: First of all, I don't understand how you got this far in the tutorial. The database was set up in chapter 4, did that work for you? The error seems to suggest you haven't upgraded your database with your migrations. Are you following the tutorial chapters in order?

Leave a Comment