2018-01-16T20:22:10Z

The Flask Mega-Tutorial Part VII: Error Handling

This is the seventh installment of the Flask Mega-Tutorial series, in which I'm going to tell you how to do error handling in a Flask application.

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

Note 1: If you are looking for the legacy version of this tutorial, it's here.

Note 2: If you would like to support my work on this blog, or just don't have patience to wait for weekly articles, I am offering the complete version of this tutorial packaged as an ebook or a set of videos. For more information, visit courses.miguelgrinberg.com.

In this chapter I'm taking a break from coding new features into my microblog application, and instead will discuss a few strategies to deal with bugs, which invariably make an appearance in any software project. To help illustrate this topic, I intentionally let a bug slip in the code that I've added in Chapter 6. Before you continue reading, see if you can find it!

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

Error Handling in Flask

What happens when an error occurs in a Flask application? The best way to find out is to experience it first hand. Go ahead and start the application, and make sure you have at least two users registered. Log in as one of the users, open the profile page and click the "Edit" link. In the profile editor, try to change the username to the username of another user that is already registered, and boom! This is going to bring a scary looking "Internal Server Error" page:

Internal Server Error

If you look in the terminal session where the application is running, you will see a stack trace of the error. Stack traces are extremely useful in debugging errors, because they show the sequence of calls in that stack, all the way to the line that produced the error:

(venv) $ flask run
 * Serving Flask app "microblog"
 * Running on http://127.0.0.1:5000/ (Press CTRL+C to quit)
[2017-09-14 22:40:02,027] ERROR in app: Exception on /edit_profile [POST]
Traceback (most recent call last):
  File "/home/miguel/microblog/venv/lib/python3.6/site-packages/sqlalchemy/engine/base.py", line 1182, in _execute_context
    context)
  File "/home/miguel/microblog/venv/lib/python3.6/site-packages/sqlalchemy/engine/default.py", line 470, in do_execute
    cursor.execute(statement, parameters)
sqlite3.IntegrityError: UNIQUE constraint failed: user.username

The stack trace indicates what is the bug. The application allows a user to change the username, and does not validate that the new username chosen does not collide with another user already in the system. The error comes from SQLAlchemy, which tries to write the new username to the database, but the database rejects it because the username column is defined with unique=True.

It is important to note that the error page that is presented to the user does not provide much information about the error, and that is good. I definitely do not want users to learn that the crash was caused by a database error, or what database I'm using, or what are some of the table and field names in my database. All that information should be kept internal.

There are a few things that are far from ideal. I have an error page that is very ugly and does not match the application layout. I also have important application stack traces being dumped on a terminal that I need to constantly watch to make sure I don't miss any errors. And of course I have a bug to fix. I'm going to address all these issues, but first, let's talk about Flask's debug mode.

Debug Mode

The way you saw that errors are handled above is great for a system that is running on a production server. If there is an error, the user gets a vague error page (though I'm going to make this error page nicer), and the important details of the error are in the server process output or in a log file.

But when you are developing your application, you can enable debug mode, a mode in which Flask outputs a really nice debugger directly on your browser. To activate debug mode, stop the application, and then set the following environment variable:

(venv) $ export FLASK_DEBUG=1

If you are on Microsoft Windows, remember to use set instead of export.

After you set FLASK_DEBUG, restart the server. The output on your terminal is going to be slightly different than what you are used to see:

(venv) microblog2 $ flask run
 * Serving Flask app "microblog"
 * Forcing debug mode on
 * Running on http://127.0.0.1:5000/ (Press CTRL+C to quit)
 * Restarting with stat
 * Debugger is active!
 * Debugger PIN: 177-562-960

Now make the application crash one more time to see the interactive debugger in your browser:

Flask Debugger

The debugger allows you expand each stack frame and see the corresponding source code. You can also open a Python prompt on any of the frames and execute any valid Python expressions, for example to check the values of variables.

It is extremely important that you never run a Flask application in debug mode on a production server. The debugger allows the user to remotely execute code in the server, so it can be an unexpected gift to a malicious user who wants to infiltrate your application or your server. As an additional security measure, the debugger running in the browser starts locked, and on first use will ask for a PIN number, which you can see in the output of the flask run command.

Since I am in the topic of debug mode, I should mention the second important feature that is enabled with debug mode, which is the reloader. This is a very useful development feature that automatically restarts the application when a source file is modified. If you run flask run while in debug mode, you can then work on your application and any time you save a file, the application will restart to pick up the new code.

Custom Error Pages

Flask provides a mechanism for an application to install its own error pages, so that your users don't have to see the plain and boring default ones. As an example, let's define custom error pages for the HTTP errors 404 and 500, the two most common ones. Defining pages for other errors works in the same way.

To declare a custom error handler, the @errorhandler decorator is used. I'm going to put my error handlers in a new app/errors.py module.

app/errors.py: Custom error handlers

from flask import render_template
from app import app, db

@app.errorhandler(404)
def not_found_error(error):
    return render_template('404.html'), 404

@app.errorhandler(500)
def internal_error(error):
    db.session.rollback()
    return render_template('500.html'), 500

The error functions work very similarly to view functions. For these two errors, I'm returning the contents of their respective templates. Note that both functions return a second value after the template, which is the error code number. For all the view functions that I created so far, I did not need to add a second return value because the default of 200 (the status code for a successful response) is what I wanted. In this case these are error pages, so I want the status code of the response to reflect that.

The error handler for the 500 errors could be invoked after a database error, which was actually the case with the username duplicate above. To make sure any failed database sessions do not interfere with any database accesses triggered by the template, I issue a session rollback. This resets the session to a clean state.

Here is the template for the 404 error:

app/templates/404.html: Not found error template

{% extends "base.html" %}

{% block content %}
    <h1>File Not Found</h1>
    <p><a href="{{ url_for('index') }}">Back</a></p>
{% endblock %}

And here is the one for the 500 error:

app/templates/500.html: Internal server error template

{% extends "base.html" %}

{% block content %}
    <h1>An unexpected error has occurred</h1>
    <p>The administrator has been notified. Sorry for the inconvenience!</p>
    <p><a href="{{ url_for('index') }}">Back</a></p>
{% endblock %}

Both templates inherit from the base.html template, so that the error page has the same look and feel as the normal pages of the application.

To get these error handlers registered with Flask, I need to import the new app/errors.py module after the application instance is created:

app/__init__.py: Import error handlers

# ...

from app import routes, models, errors

If you set FLASK_DEBUG=0 in your terminal session and then trigger the duplicate username bug one more time, you are going to see a slightly more friendly error page.

Custom 500 Error Page

Sending Errors by Email

The other problem with the default error handling provided by Flask is that there are no notifications, stack trace for errors are printed to the terminal, which means that the output of the server process needs to be monitored to discover errors. When you are running the application during development, this is perfectly fine, but once the application is deployed on a production server, nobody is going to be looking at the output, so a more robust solution needs to be put in place.

I think it is very important that I take a proactive approach regarding errors. If an error occurs on the production version of the application, I want to know right away. So my first solution is going to be to configure Flask to send me an email immediately after an error, with the stack trace of the error in the email body.

The first step is to add the email server details to the configuration file:

config.py: Email configuration

class Config(object):
    # ...
    MAIL_SERVER = os.environ.get('MAIL_SERVER')
    MAIL_PORT = int(os.environ.get('MAIL_PORT') or 25)
    MAIL_USE_TLS = os.environ.get('MAIL_USE_TLS') is not None
    MAIL_USERNAME = os.environ.get('MAIL_USERNAME')
    MAIL_PASSWORD = os.environ.get('MAIL_PASSWORD')
    ADMINS = ['your-email@example.com']

The configuration variables for email include the server and port, a boolean flag to enable encrypted connections, and optional username and password. The five configuration variables are sourced from their environment variable counterparts. If the email server is not set in the environment, then I will use that as a sign that emailing errors needs to be disabled. The email server port can also be given in an environment variable, but if not set, the standard port 25 is used. Email server credentials are by default not used, but can be provided if needed. The ADMINS configuration variable is a list of the email addresses that will receive error reports, so your own email address should be in that list.

Flask uses Python's logging package to write its logs, and this package already has the ability to send logs by email. All I need to do to get emails sent out on errors is to add a SMTPHandler instance to the Flask logger object, which is app.logger:

app/__init__.py: Log errors by email

import logging
from logging.handlers import SMTPHandler

# ...

if not app.debug:
    if app.config['MAIL_SERVER']:
        auth = None
        if app.config['MAIL_USERNAME'] or app.config['MAIL_PASSWORD']:
            auth = (app.config['MAIL_USERNAME'], app.config['MAIL_PASSWORD'])
        secure = None
        if app.config['MAIL_USE_TLS']:
            secure = ()
        mail_handler = SMTPHandler(
            mailhost=(app.config['MAIL_SERVER'], app.config['MAIL_PORT']),
            fromaddr='no-reply@' + app.config['MAIL_SERVER'],
            toaddrs=app.config['ADMINS'], subject='Microblog Failure',
            credentials=auth, secure=secure)
        mail_handler.setLevel(logging.ERROR)
        app.logger.addHandler(mail_handler)

As you can see, I'm only going to enable the email logger when the application is running without debug mode, which is indicated by app.debug being True, and also when the email server exists in the configuration.

Setting up the email logger is somewhat tedious due to having to handle optional security options that are present in many email servers. But in essence, the code above creates a SMTPHandler instance, sets its level so that it only reports errors and not warnings, informational or debugging messages, and finally attaches it to the app.logger object from Flask.

There are two approaches to test this feature. The easiest one is to use the SMTP debugging server from Python. This is a fake email server that accepts emails, but instead of sending them, it prints them to the console. To run this server, open a second terminal session and run the following command on it:

(venv) $ python -m smtpd -n -c DebuggingServer localhost:8025

Leave the debugging SMTP server running and go back to your first terminal and set export MAIL_SERVER=localhost and and MAIL_PORT=8025 in the environment (use set instead of export if you are using Microsoft Windows). Make sure the FLASK_DEBUG variable is set to 0 or not set at all, since the application will not send emails in debug mode. Run the application and trigger the SQLAlchemy error one more time to see how the terminal session running the fake email server shows an email with the full stack trace of the error.

A second testing approach for this feature is to configure a real email server. Below is the configuration to use your Gmail account's email server:

export MAIL_SERVER=smtp.googlemail.com
export MAIL_PORT=587
export MAIL_USE_TLS=1
export MAIL_USERNAME=<your-gmail-username>
export MAIL_PASSWORD=<your-gmail-password>

If you are using Microsoft Windows, remember to use set instead of export in each of the statements above.

The security features in your Gmail account may prevent the application from sending emails through it unless you explicitly allow "less secure apps" access to your Gmail account. You can read about this here, and if you are concerned about the security of your account, you can create a secondary account that you configure just for testing emails, or you can enable less secure apps only temporarily to run this test and then revert back to the default.

Logging to a File

Receiving errors via email is nice, but sometimes this isn't enough. There are some failure conditions that do not end in a Python exception and are not a major problem, but they may still be interesting enough to save for debugging purposes. For this reason, I'm also going to maintain a log file for the application.

To enable a file based log another handler, this time of type RotatingFileHandler, needs to be attached to the application logger, in a similar way to the email handler.

app/__init__.py: Email configuration

# ...
from logging.handlers import RotatingFileHandler
import os

# ...

if not app.debug:
    # ...

    if not os.path.exists('logs'):
        os.mkdir('logs')
    file_handler = RotatingFileHandler('logs/microblog.log', maxBytes=10240,
                                       backupCount=10)
    file_handler.setFormatter(logging.Formatter(
        '%(asctime)s %(levelname)s: %(message)s [in %(pathname)s:%(lineno)d]'))
    file_handler.setLevel(logging.INFO)
    app.logger.addHandler(file_handler)

    app.logger.setLevel(logging.INFO)
    app.logger.info('Microblog startup')

I'm writing the log file with name microblog.log in a logs directory, which I create if it doesn't already exist.

The RotatingFileHandler class is nice because it rotates the logs, ensuring that the log files do not grow too large when the application runs for a long time. In this case I'm limiting the size of the log file to 10KB, and I'm keeping the last ten log files as backup.

The logging.Formatter class provides custom formatting for the log messages. Since these messages are going to a file, I want them to have as much information as possible. So I'm using a format that includes the timestamp, the logging level, the message and the source file and line number from where the log entry originated.

To make the logging more useful, I'm also lowering the logging level to the INFO category, both in the application logger and the file logger handler. In case you are not familiar with the logging categories, they are DEBUG, INFO, WARNING, ERROR and CRITICAL in increasing order of severity.

As a first interesting use of the log file, the server writes a line to the logs each time it starts. When this application runs on a production server, these log entries will tell you when the server was restarted.

Fixing the Duplicate Username Bug

I have exploited the username duplication bug for too long. Now that I have showed you how to prepare the application to handle this type of errors, I can go ahead and fix it.

If you recall, the RegistrationForm already implements validation for usernames, but the requirements of the edit form are slightly different. During registration, I need to make sure the username entered in the form does not exist in the database. On the edit profile form I have to do the same check, but with one exception. If the user leaves the original username untouched, then the validation should allow it, since that username is already assigned to that user. Below you can see how I implemented the username validation for this form:

app/forms.py: Validate username in edit profile form.

class EditProfileForm(FlaskForm):
    username = StringField('Username', validators=[DataRequired()])
    about_me = TextAreaField('About me', validators=[Length(min=0, max=140)])
    submit = SubmitField('Submit')

    def __init__(self, original_username, *args, **kwargs):
        super(EditProfileForm, self).__init__(*args, **kwargs)
        self.original_username = original_username

    def validate_username(self, username):
        if username.data != self.original_username:
            user = User.query.filter_by(username=self.username.data).first()
            if user is not None:
                raise ValidationError('Please use a different username.')

The implementation is in a custom validation method, but there is an overloaded constructor that accepts the original username as an argument. This username is saved as an instance variable, and checked in the validate_username() method. If the username entered in the form is the same as the original username, then there is no reason to check the database for duplicates.

To use this new validation method, I need to add the original username argument in the view function, where the form object is created:

app/routes.py: Validate username in edit profile form.

@app.route('/edit_profile', methods=['GET', 'POST'])
@login_required
def edit_profile():
    form = EditProfileForm(current_user.username)
    # ...

Now the bug is fixed and duplicates in the edit profile form will be prevented in most cases. This is not a perfect solution, because it may not work when two or more processes are accessing the database at the same time. In that situation, a race condition could cause the validation to pass, but a moment later when the rename is attempted the database was already changed by another process and cannot rename the user. This is somewhat unlikely except for very busy applications that have a lot of server processes, so I'm not going to worry about it for now.

At this point you can try to reproduce the error one more time to see how the new form validation method prevents it.

89 comments

  • #26 srikanth said 2018-04-15T11:20:22Z

    Thanks for the great tutorial: could you explain more on below: def __init__(self, original_username, *args, **kwargs): super(EditProfileForm, self).__init__(*args, **kwargs) self.original_username = original_username this seems to be confusing.

  • #27 Miguel Grinberg said 2018-04-15T15:41:37Z

    @srikanth: this is a standard object-oriented pattern. The __init__() method first invokes the parent's class __init__() method using super(). The idea is that the derived class needs the behavior provided by the parent class, plus some new behavior. The second like implements the behavior specific to the derived class, which is to store the original_username field as an instance variable.

  • #28 YaoL said 2018-04-17T06:41:05Z

    Hi Miguel, I'm using python 3.6 on windows 10 with powershell, the code in your article where using set FLASK_DEBUG=1 doesn't work with my environment, I tried to add DEBUG = True in config.py and it also not working. Then I found that in powershell the syntax is different, I got it working with $env:FLASK_DEBUG = "1"

  • #29 Miguel Grinberg said 2018-04-17T19:31:35Z

    @YaoL: Yes, that is correct. I've made the decision of not support PowerShell in this tutorial. If you prefer to use that shell you will need minor adjustments to some commands.

  • #30 Peter said 2018-04-22T23:15:39Z

    Thank you for publishing this tutorial - it's immensely helpful. I purchased the PDF a couple of weeks back. My question with regard to the duplicate usernames/emails. The fix works, but only if the cases are the same. For example username 'Peter' is not equal to username 'peter'. Where would be the best place to address this and how would you approach it? HTML/JavaScript, form validator, etc?

  • #31 Miguel Grinberg said 2018-04-23T00:01:49Z

    @Peter: Yes, I made my comparison case sensitive. Doing case insensitive searches is possible, but it is done in different ways depending on the database engine that you use. To avoid using database specific constructs, I prefer to handle this myself, but ensuring that I always store lower cased usernames, and that I convert to lower case any username I need to search. I you wanted to do this, while at the same time preserve the original case entered by the user, you can use two columns, one for the user entered username, and another for its lower case version. You would then index the latter, so that you can do case insensitive searches.

  • #32 Harry said 2018-05-03T21:17:44Z

    I tried validating username with just the following in EditProfileForm class and did not use __init__ as shown below, but still the code works. class EditProfileForm(FlaskForm): username = StringField('Username', validators=[DataRequired()]) about_me = TextAreaField('About me', validators=[Length(min=0, max=140)]) submit = SubmitField('Submit') # def __init__(self, original_username, *args, **kwargs): # super(EditProfileForm, self).__init__(*args, **kwargs) # self.original_username = original_username def validate_username(self, username): # if username.data != self.original_username: user = User.query.filter_by(username=username.data).first() if user is not None: raise ValidationError('Username already taken!') I want to know if there are any issues by following this method.

  • #33 Miguel Grinberg said 2018-05-04T01:06:00Z

    @Harry: Your code is broken, there is no definition for self.original_username, you commented it out.

  • #34 Ogah said 2018-05-05T13:32:01Z

    Please how do I set admin e-mail in config.py. so that I can used it for authentication control

  • #35 Miguel Grinberg said 2018-05-06T05:30:14Z

    @Ogah: this is covered in this article. See the ADMINS variable in config.py.

  • #36 Theodore Grimwald said 2018-05-08T01:28:26Z

    ❤️ the many concepts, and iterative approach to your project. I've learned a lot, but its slow going for me. I found odd behavior in my follow along code, so I downloaded yours from git and am experiencing the same thing in microblog-0.7. Step 1: Register a user. Step 2: Log In with registered user. Step 3: Select "Profile" then "Edit your profile" Step 4: On the Edit Profile form enter a username that does not exist, and enter something or nothing in About me field. (get confirmation "Your changes have been saved.") Step 5 (weirdness): Select Home hyperlink... It now displays "Hi <the nonexistant user from step 4>" Step 6: Logout Step 7: Try to login with user created in Step 1, able to login in Step 2 (Invalid username or password) displayed user is no longer able to login again. Step 8: you can register the user created in step 1 again so long as you choose a different email from step 1. I don't understand why the username field would by dynamic in the edit profile, only the user themselves should be able to set/change their public information. It should be hardcoded to the current_user only. Thanks again for putting out this huge project, and making yourself available to the gaping masses (like me)

  • #37 Tiago said 2018-05-08T02:06:10Z

    @Harry, This is what you're missing: "If the user leaves the original username untouched, then the validation should allow it, since that username is already assigned to that user." In your scenario, if you try to "change" your username back to itself, your page will return an error. If you follow the code as was shown, this scenario is allowed. I have yet to find anything that has not been thoroughly explained in the tutorial (or left on purpose for later chapters). Great job!

  • #38 Theodore Grimwald said 2018-05-08T03:28:22Z

    Please, disregard my last comment (sent about two hours ago). I checked the database of the app and it actually changes the username. Able to login with new username, and the old password. Its clear to me now that was the intended functionality, chageable username. The confusion was mine. Kind regards

  • #39 Cody Reed said 2018-05-20T17:16:21Z

    Hi Miguel, Thank you so much for this awesome tutorial! I wanted to ask, I wasn't able to receive messages to the Python SMTP Debugging server running locally until I added a mail_handler formatter like the line below. What am I missing?! mail_handler.setFormatter(logging.Formatter('%(asctime)s %(levelname)s: %(message)s [in %(pathname)s:%(lineno)d]')) Using Python 2.7 alembic==0.9.9 click==6.7 Flask==1.0.2 Flask-Login==0.4.1 Flask-Migrate==2.1.1 Flask-SQLAlchemy==2.3.2 Flask-WTF==0.14.2 itsdangerous==0.24 Jinja2==2.10 Mako==1.0.7 MarkupSafe==1.0 python-dateutil==2.7.3 python-editor==1.0.3 six==1.11.0 SQLAlchemy==1.2.7 Werkzeug==0.14.1 WTForms==2.1

  • #40 Miguel Grinberg said 2018-05-20T21:52:36Z

    @Cody: Not sure what could be the problem. I haven't tested Python 2.7 in a while, but I don't think the default formatter would not work.

  • #41 Maria K said 2018-05-21T01:15:13Z

    When running the command, python -m smtpd -n -c DebuggingServer localhost:8025, should you receive a message stating that the server started? When I run it, nothing happens.

  • #42 Vassily said 2018-05-21T20:32:09Z

    Hi Miguel. For whatever reason, the mailer doesn't work for me. Guess the problem is somewhere in the settings. It's strange, as I have similar settings in Django and mail_admins(subj, msg) works just fine. Is there a way to grab the tracebacks and get them outside the logger, so I can use say Python's smtplib? Anyways, the tutorial is excellent, and thank you very much for spending your time answering the questions.

  • #43 Miguel Grinberg said 2018-05-21T21:05:21Z

    @Maria: No, I do not see any messages, the server starts silently. But when you send an email with this server you will see a dump on the console.

  • #44 Miguel Grinberg said 2018-05-22T00:02:13Z

    @Vassily: Not sure what the problem might be, but I agree that it is likely a problem with the settings. As far as implementing your own solution, you can catch errors using error handlers and implement notifications there, or you can also subclass the Flask class and override the handle_exception() method.

  • #45 Teo said 2018-05-24T11:37:47Z

    To test email you can use the mailtrap.io service.

  • #46 Aleksandr said 2018-06-04T07:37:51Z

    Hello and thank you for your tutorial, sir! I have one question. It's a bit tedious and annoying to export all these environment variables every time. And one can't do it automatically from Python script as I know. Is there way to avoid this?

  • #47 Miguel Grinberg said 2018-06-04T17:37:31Z

    @Aleksandr: you can install python-dotenv in your virtual environment and then write these variables in a .flaskenv file (dot flaskenv). Note that this requires Flask 1.0 or newer.

  • #48 Kumar said 2018-06-08T09:42:26Z

    Hello Grinberg, Thanks for your excellent tutorial. While fixing the duplicate username bug, instead of creating a constructor. How about the below code. ../forms.py from flask_login import current_user # ... def validate_username(self, username): if username.data != current_user.username: user = User.query.filter_by(username=username.data).first() if user is not None: raise ValidationError('Username already exists. Choose a different one.') The above code worked for me without any errors. Do you see any problems with this approach which I have not noticed?

  • #49 Miguel Grinberg said 2018-06-08T19:09:22Z

    @Kumar: I think your solution works as well as mine. My version of it has the advantage that it can also be used to rename users other than yourself. For example, if you wanted to have an administrator with power to rename other users, you would need to use my solution.

  • #50 Sim said 2018-06-11T13:50:55Z

    SMTP AUTH extension not supported by server. Always encounter this problem after trying gmail, live.com. Nightmare lOl

Leave a Comment