This directory contains automated integration tests for Cockpit, and the support files for them. The architecture of the automated integration tests is described in ARCHITECTURE
To run the tests on Fedora, refer to the HACKING guide for installation of all of the necessary build and test dependencies. There's no need to trigger a build manually - the test suite preparation step below will handle that.
If test failures are encountered that look like they may be related to problems with nested virtualization, refer to this Fedora guide for more details and recommendations on ensuring it is enabled correctly.
Warning: Never run the build, test, or any other command here as root!
You first need to build cockpit, and install it into a VM:
test/image-prepare
This uses the default OS image. See $TEST_OS below how to select a different one.
See test/image-prepare --help for some special modes, like skipping unit
tests, building overlays, or preparing an image with the
cockpit/ws container instead of RPMs.
See our scenarios in test/run for how they are being used.
In most cases you want to run an individual test in a suite, for example:
test/verify/check-metrics TestCurrentMetrics.testCPU
You can get a list of tests by inspecting the def test* in the source, or by
running the suite with -l/--list:
test/verify/check-metrics -l
Sometimes you may also want to run all tests in a test file suite:
test/verify/check-session
To see more verbose output from the test, use the -v/--verbose and/or -t/--trace flags:
test/verify/check-session --verbose --trace
If you specify -s/--sit in addition, then the test will wait on failure and
allow you to log into cockpit and/or the test instance and diagnose the issue.
The cockpit and SSH addresses of the test instance will be printed:
test/verify/check-session -st
You can also run all the tests, with some parallelism:
test/common/run-tests --test-dir test/verify --jobs 2
However, this will take really long. You can specify a subset of tests (see
--help); but usually it's better to run individual tests locally, and let the
CI machinery run all of them in a draft pull request.
The tests will automatically download the VM images they need, so expect that the initial run may take a few minutes.
Normally each test starts its own chromium headless browser process on a separate random port. To interactively follow what a test is doing:
TEST_SHOW_BROWSER=1 test/verify/check-session --trace
You can also run a test against Firefox instead of Chromium:
TEST_BROWSER=firefox test/verify/check-session --trace
See below for details.
You can conduct manual interactive testing against a test image by starting the image like so:
bots/vm-run -s cockpit.socket debian-stable
Once the machine is booted and the cockpit socket has been activated, a message will be printed describing how to access the virtual machine, via ssh and web. See the "Helpful tips" section below.
By default, it's only possible to contact the virtual machine from the host
machine on which it's running. If you want to conduct manual testing from
other devices on your network, set TEST_BIND_GLOBAL=1, for example:
TEST_BIND_GLOBAL=1 bots/vm-run -s cockpit.socket debian-stable
This will bind the Cockpit and SSH ports to all interfaces, making it possible to access a URL like http://yourhost.local:9091/ to test Cockpit from another machine on your LAN.
Pixel tests in Cockpit ensure that updates of our dependencies or code changes don't break the UI: for example slight changes of layout, padding, color and everything which isn't easily spotted by a human. They also give us confidence that an update of our UI Framework doesn't introduce changes in how Cockpit looks.
Pixel tests make a screenshot of a selector and compare it to a known good reference image. if there is a difference, the test fails and a pixel difference is shown.
This works as our tests run in the cockpit/tasks container which pins the browser and font rendering so repeated runs provide the same pixels. To generate new pixels, this tasks container must be used; your own browser and font rendering software might generate different results. For more information read the "introduction blog post".
The test images are stored in a git submodule in the test/reference directory
and be fetched with:
./test/common/pixel-tests updateAs Cockpit tests under multiple distributions and it is not worth the effort to
run pixel tests on every supported distribution we only run them for the
image configured in test/reference-image.
Our tests call Browser.assert_pixels at interesting and strategic places.
This assertion method requires at least a CSS selector and an image title.
Pixel tests are generated in five layouts by default: desktop, medium, mobile,
dark and rtl.
Take a screenshot of the content in #detail-content:
browser.assert_pixels("#detail-content", "filesystem")Take a screenshot of the content in #detail-content and ignore all elements
with a class disk-stats as they change per test run:
browser.assert_pixels("#detail-content", "filesystem", ignore=[".disks-stats"])Take a screenshot of the content in #detail-content and skip it for a
specific layout as it generates unstable pixels:
browser.assert_pixels("#detail-content", "filesystem", skip_layouts=["rtl"])To update pixel tests, locally run the test in the current tasks container, or
create a draft PR and let the tests run for test/reference-image and
afterwards fetch the new pixels:
./test/common/pixel-tests fetch "https://cockpit-logs.us-east-1.linodeobjects.com/<snip>/log.html"
Finally, upload the new pixel tests and commit the newly generated submodule commit:
./test/common/pixel-tests push
Note that you have to a part of the Contributors group to push pixel tests.
You can set these environment variables to configure the test suite:
-
TEST_OS: The OS to run the tests in, like "fedora-coreos" or "debian-stable". See the "cockpit-project/cockpit" section in the test map for all supported values. "fedora-40" is the default (TEST_OS_DEFAULTin bots' constants.py). -
TEST_JOBS: How many tests to run in parallel. The default is 1. -
TEST_BROWSER: What browser should be used for testing. Currently supported values are "chromium" and "firefox". "chromium" is the default. -
TEST_SHOW_BROWSER: Set to run browser interactively. When not specified, browser is run in headless mode. When set to "pixels", the browser will be resized to the exact dimensions that are used for pixel tests. -
TEST_TIMEOUT_FACTOR: Scale normal timeouts by given integer. Useful for slow/busy testbeds or architectures.
See the bots documentation for details about the tools and configuration for these.
It is recommended to add a snippet like this to your ~/.ssh/config. Then
you can log in to test machines without authentication:
Match final host 127.0.0.2
User root
StrictHostKeyChecking no
UserKnownHostsFile /dev/null
CheckHostIp no
IdentityFile CHECKOUT_DIR/bots/machine/identity
IdentitiesOnly yes
You need to replace CHECKOUT_DIR with the actual directory where you cloned
cockpit.git, or bots.git if you have a separate clone for that.
Many cockpit developers take it a step further, and add an alias to
allow typing ssh c:
Host c
Hostname 127.0.0.2
Port 2201
The final keyword in the first rule will cause it to be checked (and matched)
after the Hostname substitution in the c rule.
Each image-prepare invocation will always start from the pristine image and
ignore the current overlay in test/images. It is thorough, but also rather
slow. If you want to iterate on changing only JavaScript/HTML code, as opposed
to the bridge or webserver, the whole build and test cycle can be done much
faster.
You always need to do at least one initial test/image-prepare $TEST_OS run.
Afterwards it depends on the kind of test you want to run.
Many test methods or classes are marked as @nondestructive, meaning that
they restore the state of the test VM enough that other tests can run
afterwards. This is the fastest and most convenient situation for both
iterating on the code and debugging failing tests.
Start the prepared VM with bots/vm-run $TEST_OS. Note the SSH and cockpit
ports. If this is the only running VM, it will have the ports in the
examples below, otherwise the port will be different.
Then start building the page you are working on in watch and rsync mode, e.g.
RSYNC=c ./build.js -w users
(Assuming the c SSH alias from the previous section and first running VM).
Then you can run a corresponding test against the running VM, with additional debug output:
TEST_OS=... test/verify/check-users -t --machine 127.0.0.2:2201 --browser 127.0.0.2:9091 TestAccounts.testBasic
Other tests need one or more fresh VMs. Instead of a full test/image-prepare
run (which is slow), you can update the existing VM overlay with updated
bundles. Start the build in watch mode, but without rsyncing, e.g.
./build.js -w storaged
and after each iteration, copy the new bundles into the VM overlay:
bots/image-customize -u dist:/usr/share/cockpit/ $TEST_OS
Then run the test as you would normally do, e.g.
TEST_OS=... test/verify/check-storage-stratis -t TestStorageStratis.testBasic
Use bots/vm-reset to clean up all prepared overlays in test/images.
If you pass the -s ("sit on failure") option to a test program, it
will pause when a failure occurs so that you can log into the test
machine and investigate the problem.
A test will print out the commands to access it when it fails in this way. You can log into a running test-machine using ssh. See the "Helpful tips" section below.
You can also put calls to sit() into the tests themselves to stop them
at strategic places.
That way, you can run a test cleanly while still being able to make quick changes, such as adding debugging output to JavaScript.
If a test is not decorated with @nondestructive, it is OK for a test to
destroy the test machine OS installation, or otherwise modify it without
cleaning up. For example, it is OK to remove all of /etc just to see what
happens. The next test will get a pristine test machine.
Tests decorated with @nondestructive will all run against the same test
machine. The nondestructive test should clean up after itself and restore the
state of the machine, such that the next nondestructive test is not impacted.
A fast running test suite is more important than independent, small test cases.
Thus, it is OK for tests to be long. Starting the test machine is so slow that we should run as many checks within a single session as make sense. Note that nondestructive tests do not suffer from this, and are much quicker.
Still, within a long test, try to have independent sections, where each section returns the machine to more or less the state that it was in before the section. This makes it easier to run these sections ad-hoc when doing incremental development.
Every pull request will trigger a $DEFAULT_OS/devel scenario which creates a
coverage report of the JavaScript code executed and writes comments about
uncovered code in the pull request. The overall coverage percentage is recorded
in prometheus for a subset of our projects and visualized in Grafana.
To generate coverage locally for TestApps:
export NODE_ENV=devel
./build.js
./test/image-prepare -q
./test/common/run-tests --test-dir test/verify --coverage TestApps
Code which is impossible or very hard to test in our tests can be excluded from
appearing in a pull request as comment by adding a not-covered comment with a
short justification:
return cockpit.script(data, { superuser: "try", err: "message" })
.catch(console.error); // not-covered: OS errorFor web access, if you'd like to avoid Chromium (or Chrome) prompting about certificate errors while connecting to localhost, you can change the following setting:
chrome://flags/#allow-insecure-localhost