Dash Testing

New in Dash v1.0
support Dash for R testing added in v1.1.0

dash.testing provides some off-the-rack pytest fixtures and a minimal set of testing APIs with our internal crafted best practices at the integration level. This tutorial does not intend to cover the usage of pytest and Selenium WebDriver, but focuses on how to do a simple integration test with Dash by hosting the App server locally and using a Selenium WebDriver to simulate the interaction inside a web browser.

Install

The Dash testing is now part of the main Dash package. After
pip install dash[testing], the Dash pytest fixtures are available, you
just need to install the WebDrivers or use a remote Selenium-Grid and you
are ready to test.
- Chrome Driver
- Firefox Gecko Driver
FYI, We run Dash integration tests with Chrome WebDriver.
But the fixture allows you to choose another browser from the command line,
e.g. pytest --webdriver Firefox -k bsly001.
Headless mode is added in Dash 1.0.1, run pytest --headless -k bsly001
to start test in headless mode. First time hearing about headless mode? The
main benefit for us is it’s lighter and faster to run without a UI. You
can check the details from both Firefox
and Chrome.
Remote WebDriver support is added in Dash 1.3.0. There are two ways to use it:
1. Run pytest --remote -k bsly001 to grab a Chrome WebDriver from a local
hosted grid at http://localhost:4444/wd/hub
2. Run pytest --webdriver Firefox --remote-url <a href="https://grid_provioder_endpoints“>https://grid_provioder_endpoints</a> to connect with a remote grid in the cloud running Firefox (default Chrome). Note that you don't need to use–remoteas soon as the–remote-url`
value is set and different than the default one.

Caveats

It’s important to note that we cannot fully test and guarantee that the
above cases will work with any given selenium grid you have. The limitation
might come from how the network is set up, the limitation of different
hosting OS or how docker-compose was configured.
You might need to do some auxiliary WebDriver Options tuning to run the
tests in a particular Selenium-Grid. The first useful tip is to change the
default logging level with --log-cli-level DEBUG. Secondly, there is a
back door for browser option customization by a pytest_setup_options hook
defined in plugin.py.
The example below is to use the headless mode with Chrome WebDriver in
Windows, there is a workaround
by adding --disable-gpu in the options.

# add this in the conftest.py under tests folder
from selenium.webdriver.chrome.options import Options

def pytest_setup_options():
    options = Options()
    options.add_argument('--disable-gpu')
    return options

Notes: * The Gecko(Marionette) driver from Mozilla is not fully compatible with
Selenium specifications. Some features may not work as expected. * We only include Chrome and Firefox in the supported list for now,
but other popular webdrivers may be included based on popular demand.

Example

# 1. imports of your dash app
import dash
from dash import html
# 2. give each testcase a tcid, and pass the fixture
# as a function argument, less boilerplate
def test_bsly001_falsy_child(dash_duo):
    # 3. define your app inside the test function
    app = dash.Dash(__name__)
    app.layout = html.Div(id="nully-wrapper", children=0)
    # 4. host the app locally in a thread, all dash server configs could be
    # passed after the first app argument
    dash_duo.start_server(app)
    # 5. use wait_for_* if your target element is the result of a callback,
    # keep in mind even the initial rendering can trigger callbacks
    dash_duo.wait_for_text_to_equal("#nully-wrapper", "0", timeout=4)
    # 6. use this form if its present is expected at the action point
    assert dash_duo.find_element("#nully-wrapper").text == "0"
    # 7. to make the checkpoint more readable, you can describe the
    # acceptance criterion as an assert message after the comma.
    assert dash_duo.get_logs() == [], "browser console should contain no error"
    # 8. visual testing with percy snapshot
    dash_duo.percy_snapshot("bsly001-layout")

Notes

1. For most test scenarios, you don’t need to import any modules for
   the test; just import what you need for the Dash app itself.
2. A test case is a regular Python function. The function name follows
   this pattern: test_{tcid}_{test title}. The tcid (test case ID) is
   an abbreviation pattern of mmffddd => module + file + three digits.
   The tcid facilitates the test selection by just running
   pytest -k {tcid}. Its naming convention also helps code navigation with
   modern editors.
3. Here we just define our app inside a test function. All the rules
   still apply as in your app file.
4. We normally start the test by calling the start_server API
   from dash_duo (you can use dash_br for an hosted Dash App, and
   write dash_br.server_url = "Hosted URL" to start). Several actions
   implicitly happen under the hood:
5. The defined app is hosted inside a light Python threading.Thread.
6. A Selenium WebDriver is initialized and navigates to the
   local server URL using server_url.
7. We first wait until the Flask server is responsive to an HTTP
   request; and then make sure the Dash app is full rendered inside
   the browser.
8. A test case is composed of preparation, actions, and checkpoints.
   Both #5 and #6 are doing the same check in this example; we are expecting
   that the defined Div component’s text is identical to children. #5 will
   wait for the expected state to be reached within a 4 seconds timeout. It’s
   a safer way to write the action steps when you are doing an element check
   related to callbacks, as it normally happens under Dash context:
   the element is already present in the DOM, but not necessarily the props.
9. The find_element API call has an implicitly global timeout of two
   seconds set at the driver level, i.e. the driver waits at most two seconds
   to find the element by the locator, HOWEVER it will compare the text
   as soon as the driver returns an element. Also note that the API
   find_element('#nully-wrapper') is just a shortcut to a more tedious
   version driver.find_element_by_css_selector('#nully-wrapper').
10. Unlike unittest, pytest allows using the standard Python
    assert for verifying
    expectations and values. It also puts more introspection information into
    the assertion failure message by overriding the assert behavior. It’s
    good practice to expose your acceptance criteria directly in the test
    case rather than wrapping the assert inside another helper API, also to
    write these messages with SHOULD/SHOULD NOT without failure confusion.
    By looking at the test name, the app definition, the actions, and the
    checkpoints, reviewers should figure out easily the purpose of the test.
11. We use Percy as our Visual Regression Testing
    tool. It’s a good alternative to assertions when your checkpoint is
    about the graphical aspects of a Dash App, such as the whole layout or a
    dcc.Graph component. We integrate the Percy service with a PERCY_TOKEN
    variable, so the regression result is only available in Plotly’s CircleCI
    setup.

Dash for R Testing

We released Dash for R
in July 2019. To facilitate testing it, we extended the Python package
dash.testing to support Dash for R apps.
The new dashr fixture gives us a test development experience in
Dash for R that’s nearly identical to the dash_duo fixture in
Dash for Python. In this context, a dashr fixture is an instance of
Python class which provides a reliable framework to execute the app and test
one or more of its features via Selenium WebDriver. For more details,
please visit pytest documentation.
Here is a simple example runnable with pytest. Your Dash App is written
as a string inside the Python test code (the app argument can also be a
valid path to a .R file), the app is then executed by Rscript within
a Python subprocess and we can use the same set of API calls to
test it exactly as we do Dash testing in Python Apps.

# the app is a raw string variable defining the Dash App in R
app = '''
library(dash)
library(dashHtmlComponents)
app <- Dash$new()
app$layout(htmlDiv(list(htmlDiv(id='container',children='Hello Dash for R testing'))))
app$run_server()
'''
# a test case is a simple Python function with the same prefix convention
# dashr is the default fixture combines the API for serving the app
# and selenium tests.
def test_rstr001_r_with_string(dashr):
    # Since the app is the string chunk we defined above, the dashr
    # fixture creates an unique temporary folder to dump the content into
    # `app.R` and copies any possible subfolders (usually
    # assets or clientside source code) to the same temporary folder
    # start_server by default uses the temporary folder as current working
    # folder You can change this with `start_server(app, cwd='/my/cwd/')`
    dashr.start_server(app)
    assert dashr.find_element("#container").text == "Hello Dash for R Testing"

def test_rstr002_r_with_file_path(dashr):
    # Alternatively, the app can be a path to a separate file containing your Dash for R app
    # the `os.path.dirname(__file__)` is an useful Python trick to get the path
    # of the currently running script, so it will always work no matter where you start the
    # test runner.
    dashr.start_server(
        app=os.path.join(os.path.dirname(__file__), "assets/demo_hello.R))
    assert dashr.find_element("#container").text == "Hello Dash for R Testing"

Fixtures

To avoid accidental name collision with other pytest plugins, all Dash test
fixtures start with the prefix dash or dashr.
- dash_br
A standalone WebDriver wrapped with high-level Dash testing APIs. This is
suitable for testing a Dash App in a deployed environment (Dash for Python
or R), i.e. when your Dash App is accessible from a URL.
- dash_duo
The default fixture for Dash Python integration tests, it contains a
thread_server and a WebDriver wrapped with high-level Dash testing APIs.
- dash_thread_server
Start your Dash App locally in a Python threading.Thread, which is
lighter and faster than a process.
- dash_process_server
This is close to your production/deployed environment. Start your Dash App
with waitress(by default if raw_command is not provided) in a Python
subprocess. You can control the process runner with two supplemental
arguments. To run the application with alternative deployment options, use
the raw_command argument; to extend the timeout if your application needs
more than the default three seconds to launch, use the start_timeout
argument. Note: You need to configure your PYTHONPATH so that the Dash
app source file is directly importable.
And Dash for R test fixtures have a prefix dashr.
- dashr
The default fixture for Dash for R integration tests. As dash_duo in Python,
it contains a dashr_server and a Selenium WebDriver with the same Dash
testing APIs.
- dashr_server
Start your Dash for R App with Rscript in a python subprocess. Note that
unlike python server fixtures, the start server arguments can only be
configured inside the Dash for R App.

APIs

Selenium Overview

Both dash_duo and dash_br expose the Selenium WebDriver via the
property driver, e.g. dash_duo.driver, which gives you full access to
the Python Selenium API.
(Note that this is not the official Selenium documentation site, but has
somehow become the defacto Python community reference)
One of the core components of selenium testing is finding the
web element with a locator, and performing some actions like click
or send_keys on it, and waiting to verify if the expected state is met
after those actions. The check is considered as an acceptance criterion,
for which you can write in a standard Python assert statement.

Element Locators

There are several strategies to
locate elements;
CSS selector and XPATH are the two most versatile ways. We recommend using
the CSS Selector in most cases due to its
better performance and robustness across browsers.
If you are new to using CSS Selectors, these
SauceLab tips
are a great start. Also, remember that
Chrome Dev Tools Console
is always your good friend and playground.

Waits

This link covers
this topic nicely. For impatient readers, a quick take away is
quoted as follows:
The Selenium WebDriver provides two types of waits:
- explicit wait
Makes WebDriver wait for a certain condition to occur before
proceeding further with execution. All our APIs with wait_for_*
falls into this category.
- implicit wait
Makes WebDriver poll the DOM for a certain amount of time when trying
to locate an element. We set a global two-second timeout at the
driver level.
Note all custom wait conditions are defined in dash.testing.wait
and there are two extra APIs until and until_not which are similar to
the explicit wait with WebDriver, but they are not binding to
WebDriver context, i.e. they abstract a more generic mechanism to
poll and wait for certain condition to happen

Browser APIs

This section lists a minimal set of browser testing APIs. They are
convenient shortcuts to Selenium APIs and have been approved in
our daily integration tests.
The following table might grow as we start migrating more legacy tests in
the near future. But we have no intention to build a comprehensive list,
the goal is to make writing Dash tests concise and error-free.
Please feel free to submit a community PR to add any missing ingredient,
we would be happy to accept that if it’s adequate for Dash testing.

Dash APIs

This section enumerates a full list of Dash App related properties and APIs
apart from the previous browser ones.

Debugging

Verify Your Test Environment

If you run the integration in a virtual environment, make sure you are
getting the latest commit in the master branch from each component, and
that the installed pip versions are correct.
Note: We have some enhancement initiatives tracking in this issue

Run the CI Job Locally

The CircleCI Local CLI is a
handy tool to execute some jobs locally. It gives you an earlier warning
before even pushing your commits to remote. For example, it’s always
recommended to pass lint and unit tests job first on your local machine. So
we can make sure there are no simple mistakes in the commit.

# install the cli (first time only)
$ curl -fLSs <a href="https://circle.ci/cli">https://circle.ci/cli</a> | bash && circleci version
# run at least the lint & unit test job on both python 2 and 3
# note: the current config requires all tests pass on python 2.7, 3.6 and 3.7.
$ circleci local execute --job lint-unit-27 && $ circleci local execute --job lint-unit-37

Increase the Verbosity of pytest Logging Level

pytest --log-cli-level DEBUG -k bsly001

You can get more logging information from Selenium WebDriver, Flask server,
and our test APIs.

14:05:41 | DEBUG | selenium.webdriver.remote.remote_connection:388 | DELETE <a href="http://127.0.0.1:53672/session/87b6f1ed3710173eff8037447e2b8f56">http://127.0.0.1:53672/session/87b6f1ed3710173eff8037447e2b8f56</a> {"sessionId": "87b6f1ed3710173eff8037447e2b8f56"}
14:05:41 | DEBUG | urllib3.connectionpool:393 | <a href="http://127.0.0.1:53672">http://127.0.0.1:53672</a> "DELETE /session/87b6f1ed3710173eff8037447e2b8f56 HTTP/1.1" 200 72
14:05:41 | DEBUG | selenium.webdriver.remote.remote_connection:440 | Finished Request
14:05:41 | INFO | dash.testing.application_runners:80 | killing the app runner
14:05:41 | DEBUG | urllib3.connectionpool:205 | Starting new HTTP connection (1): localhost:8050
14:05:41 | DEBUG | urllib3.connectionpool:393 | <a href="http://localhost:8050">http://localhost:8050</a> "GET /_stop-3ef0e64e8688436caced44e9f39d4263 HTTP/1.1" 200 29

Selenium Snapshots

If you run your tests with CircleCI dockers (locally with CircleCI CLI
and/or remotely with CircleCI).
Inside a docker run or VM instance where there is no direct access to the
video card, there is a known limitation that you cannot see anything from
the Selenium browser on your screen. Automation developers use
Xvfb as
a workaround to solve this issue. It enables you to run graphical
applications without a display (e.g., browser tests on a CI server) while
also having the ability to take screenshots.
We implemented an automatic hook at the test report stage, it checks if a
test case failed with a Selenium test fixture. Before tearing down every
instance, it will take a snapshot at the moment where your assertion is
False or having a runtime error. refer to Browser APIs
Note: you can also check the snapshot directly in CircleCI web page
under Artifacts Tab

Percy Snapshots

There are two customized pytest arguments to tune Percy runner:
1. --nopercyfinalize disables the Percy finalize in dash fixtures. This
is required if you run your tests in parallel, then you add an extra
percy finalize --all step at the end. For more details, please visit
Percy Documents.
2. --percy-assets lets Percy know where to collect additional assets
such as CSS files. You can refer to the example we used for dash-docs.