django-smoketest

PLEASE NOTE: I'm basically doing README-driven development here, writing documentation for how this code should work before actually implementing it. This notice will go away when django-smoketest is actually implemented and remotely suitable for real-world use. Until then, feel free to offer ideas on the interface, but don't expect to be able to use it (you can look in the "Progress" section to see exactly where I'm at).

Motivation

Smoke test framework for Django.

Smoke tests are tests that are run on a production environment to quickly detect major systemic problems. Eg, after you run a deploy, you want to quickly check that everything is running properly so you can roll back quickly instead if there are problems. Too often, this just means visiting the site and manually clicking around through a few links (at best).

You probably already have unit tests verifying the correctness of low level parts of your code, and integration and acceptance tests running on a staging server or CI system. Maybe you've even got automatic configuration management ensuring that your staging server is configured as an exact replica of production. So logically, if your code passes all the tests on the staging server and the production server is configured the same, everything must work right in production. Right? Wouldn't it be wonderful if the world were so simple? Of course we know that it's not. That's why we want smoke tests to actually verify that at least the major components of the system are all basically functional and able to talk to each other and we didn't do something stupid like writing code that depends on a new environment variable that hasn't been set to the correct value on production yet.

You probably don't want to run your unit tests or integration tests in production with production settings in effect. Who knows what kind of insanity would result? Test data sprayed all through your production database, deleting user data from the file system, the sun rising in the west and setting in the east?

This is what smoke tests are for. Smoke tests should be safe to run in production. Verify that the application can connect to the database, that whatever filesystem mounts are expected are in place, etc. bridging that last gap between existing test coverage and the wilderness of production. But all while stepping carefully around the production data.

I also find myself frequently writing small views to support ad-hoc monitoring. Eg, if an application relies on an NFS mount for some infrequent operation and that mount has a tendency to go stale, a cron job that runs every few minutes (or via nagios or some other monitoring application) and has the application try to read a file off the mount can help ensure that we are alerted to the stale mount before users encounter it.

Getting Started

Install django-smoketest

$ pip install django-smoketest

Add smoketest to your INSTALLED_APPLICATIONS.

In each application of yours that you want to define smoke tests for, make a smoke.py file or a smoke directory with an __init__.py and one or more python files with your tests.

In your urls.py, add something like:

('smoketest/', include('smoketest.urls'))

To your urlpatterns.

In your smoke.py (or module), you put something like this:

from smoketest import SmokeTest
from smoketest.decorators import slow, rolled_back
from myapp.models import FooModel


class DemoTest(SmokeTest):
    def test_foomodel_reads(self):
        """ just make sure we can read data from the db """
        cnt = FooModel.objects.all().count()
        self.assertTrue(cnt > 0)

    @rolled_back
    def test_foomodel_writes(self):
        """ make sure we can also write to the database
        but do not leave any test detritus around.
        """
        f = FooModel.objects.create()
    
    @slow
    def test_something_slow(self):
        """ this test will not be run in "fast" mode
        because it uses a lot of resources or otherwise
        could bog down the production server in bad ways
        """
        # do a bunch of slow stuff
        # ...
        self.assertEqual(foo, bar)

Now, if you make a GET to http://yourapp/smoketest/, django-smoketest will go through your code, finding any smoke modules, and run the tests you have defined (if you've used unittest or nose, you get the idea), excluding any marked with the @slow decorator. GETing http://yourapp/smoketest/slow/ will include those tests as well. All tests passing will result in a response like:

PASS
test classes: 1
tests run: 3
tests passed: 3
tests failed: 0
tests errored: 0
time: 1200.307861328ms

So you can just check the result for PASS if you are calling it from a monitoring script or as part of an automated deploy.

If tests fail or error out, you instead get something like:

FAIL
test classes: 1
tests run: 8
tests passed: 5
tests failed: 2
tests errored: 1
time: 3300.07861328ms
module1.smoke.DemoTest.test_foo failed
module1.smoke.DemoTest.test_bar failed
module1.smoke.DemoTest.test_baz errored

If your HTTP client makes the request with application/json in the Accept: headers, responses will be JSON objects with the same information in a more easily parseable form:

$ curl -H "Accept: application/json" http://yourapp/smoketest/
{"status": "FAIL", "tests_failed": 2,
 "errored_tests": ["module1.smoke.DemoTest.test_baz"],
 "tests_run": 8, "test_classes": 1, "tests_passed": 5,
 "failed_tests": ["module1.smoke.DemoTest.test_foo",
 "module1.smoke.DemoTest.test_foo"], "tests_errored": 1,
 "time": 1.6458759307861328}

QUESTION: I'm thinking about keeping the output simple to parse automatically, but maybe we ought to just stick with unittest's existing output format instead?

API

The main class is smoketests.SmokeTest, which should be though of as equivalent to unittest.TestCase. It will do basically the usual stuff there, running setUp and tearDown methods, and supporting the usual array of assertEquals, assertRaises, assertTrue methods.

There is the @slow decorator which marks a test as potentially slow, or utilizing a lot of resources. Either way, it lets you have two different levels of smoke tests. Fast tests can be run frequently, eg, from a monitoring script that hits it every five minutes so you can quickly be alerted if something changes in the production environment. The @slow tests can then be reserved for only running after a new deploy to check things a little more deeply and have more confidence that everything is functional.

The @rolled_back decorator will make sure that the test gets wrapped in a database transaction which is then rolled back after running. This frees you up to do potentially destructive things and just let the DB clean up for you. The usual caveats apply about making sure you are using a database that supports transactions and that it can only roll back database operations, not other side effects. I'm also on the fence about whether this decorator should even exist or if that should be the default behavior for all smoke tests. Should a smoke test ever actually commit a transaction?

In your settings, you may define a SMOKETEST_APPS variable that lists the applications want to run smoke tests from (instead of looking through all your applications). (do we want a SMOKETEST_SKIP_APPS as well/instead?).

Asserts supported (so far):

• assertEqual(a, b)
• assertNotEqual(a, b)
• assertTrue(t)
• assertFalse(x)
• assertIs(a, b)
• assertIsNot(a, b)
• assertIsNone(x)
• assertIsNotNone(x)
• assertIn(a, b)
• assertNotIn(a, b)
• assertIsInstance(a, b)
• assertNotIsInstance(a, b)

Open Questions

What other unittest/nose flags, conventions, etc should we support? --failfast? output verbosity? ability to target or skip specific tests in certain cases? Automatic timeouts (a lot of smoke tests involve trying to connect to an external service and failing if it takes more than a specified period of time)?

Progress

TODO:

• @slow decorator and view
• @rolled_back decorator
• capture stdout/stderr
• I think it only handles smoke.py files or smoke/__init__.py and won't yet find subclasses in submodules like smoke/foo.py.
• report additional info (exception/tracebacks) on errors
• support messages on asserts
• setUpClass/tearDownClass
• extended assert* methods (listed in smoketest/__init__.py)
• SMOKETEST_APPS (and/or SMOKETEST_SKIP_APPS)

DONE:

• walk INSTALLED_APPLICATIONS and find/run smoke tests
• report numbers in simple text format
• run setUp and tearDown methods
• when tests fail/error, report which ones failed/errored
• proper module.class.method info on test failures/errors report
• support the basic expected set of assert* methods from unittest
• JSON output
• time test runs and include in output
• run tests in a rolled back transaction