Unit and Integration Tests
Starting with Qubes R3 we use python unittest to perform automatic tests of Qubes OS. Despite the name, we use it for both unit tests and integration tests. The main purpose is, of course, to deliver much more stable releases.
Integration tests are written with the assumption that they will be called on dedicated hardware.
Do not run these tests on installations with important data, because you might lose it.
Since these tests were written with this expectation, all the VMs with a name starting with
test- on the installation are removed during the process, and all the tests are recklessly started from dom0, even when testing VM components.
Most of the tests are stored in the core-admin repository in the
To start them you can use standard python unittest runner:
python3 -m unittest -v qubes.tests
Or our custom one:
python3 -m qubes.tests.run -v
Our test runner runs mostly the same as the standard one, but it has some nice additional features like color output and not needing the “qubes.test” prefix. It also has the ability to run lone selected template tests.
You can use
python3 -m qubes.tests.run -h to get usage information:
[user@dom0 ~]$ python3 -m qubes.tests.run -h usage: run.py [-h] [--verbose] [--quiet] [--list] [--failfast] [--no-failfast] [--do-not-clean] [--do-clean] [--loglevel LEVEL] [--logfile FILE] [--syslog] [--no-syslog] [--kmsg] [--no-kmsg] [TESTNAME [TESTNAME ...]] positional arguments: TESTNAME list of tests to run named like in description (default: run all tests) optional arguments: -h, --help show this help message and exit --verbose, -v increase console verbosity level --quiet, -q decrease console verbosity level --list, -l list all available tests and exit --failfast, -f stop on the first fail, error or unexpected success --no-failfast disable --failfast --loglevel LEVEL, -L LEVEL logging level for file and syslog forwarding (one of: NOTSET, DEBUG, INFO, WARN, WARNING, ERROR, CRITICAL; default: DEBUG) --logfile FILE, -o FILE if set, test run will be also logged to file --syslog reenable logging to syslog --no-syslog disable logging to syslog --kmsg, --very-brave-or-very-stupid log most important things to kernel ring-buffer --no-kmsg, --i-am-smarter-than-kay-sievers do not abuse kernel ring-buffer --allow-running-along-qubesd allow running in parallel with qubesd; this is DANGEROUS and WILL RESULT IN INCONSISTENT SYSTEM STATE --break-to-repl break to REPL after tests When running only specific tests, write their names like in log, in format: MODULE+"/"+CLASS+"/"+FUNCTION. MODULE should omit initial "qubes.tests.". Example: basic/TC_00_Basic/test_000_create
For instance, to run only the tests for the fedora-21 template, you can use the
-l option, then filter the list:
[user@dom0 ~]$ python3 -m qubes.tests.run -l | grep fedora-21 network/VmNetworking_fedora-21/test_000_simple_networking network/VmNetworking_fedora-21/test_010_simple_proxyvm network/VmNetworking_fedora-21/test_020_simple_proxyvm_nm network/VmNetworking_fedora-21/test_030_firewallvm_firewall network/VmNetworking_fedora-21/test_040_inter_vm vm_qrexec_gui/TC_00_AppVM_fedora-21/test_000_start_shutdown vm_qrexec_gui/TC_00_AppVM_fedora-21/test_010_run_gui_app vm_qrexec_gui/TC_00_AppVM_fedora-21/test_050_qrexec_simple_eof vm_qrexec_gui/TC_00_AppVM_fedora-21/test_051_qrexec_simple_eof_reverse vm_qrexec_gui/TC_00_AppVM_fedora-21/test_052_qrexec_vm_service_eof vm_qrexec_gui/TC_00_AppVM_fedora-21/test_053_qrexec_vm_service_eof_reverse vm_qrexec_gui/TC_00_AppVM_fedora-21/test_060_qrexec_exit_code_dom0 vm_qrexec_gui/TC_00_AppVM_fedora-21/test_065_qrexec_exit_code_vm vm_qrexec_gui/TC_00_AppVM_fedora-21/test_100_qrexec_filecopy vm_qrexec_gui/TC_00_AppVM_fedora-21/test_110_qrexec_filecopy_deny vm_qrexec_gui/TC_00_AppVM_fedora-21/test_120_qrexec_filecopy_self vm_qrexec_gui/TC_20_DispVM_fedora-21/test_000_prepare_dvm vm_qrexec_gui/TC_20_DispVM_fedora-21/test_010_simple_dvm_run vm_qrexec_gui/TC_20_DispVM_fedora-21/test_020_gui_app vm_qrexec_gui/TC_20_DispVM_fedora-21/test_030_edit_file [user@dom0 ~]$ python3 -m qubes.tests.run -v `python3 -m qubes.tests.run -l | grep fedora-21`
Example test run:
Tests on Qubes 4.0 require stopping
qubesd service first, because special instance of it is started as part of the test run.
Additionally, tests needs to be started as root. The full command to run the tests is:
sudo systemctl stop qubesd; sudo -E python3 -m qubes.tests.run -v ; sudo systemctl start qubesd
On Qubes 4.0 tests are also compatible with nose2 test runner, so you can use this instead:
sudo systemctl stop qubesd; sudo -E nose2 -v --plugin nose2.plugins.loader.loadtests qubes.tests; sudo systemctl start qubesd
This may be especially useful together with various nose2 plugins to store tests results (for example
nose2.plugins.junitxml), to ease presenting results. This is what we use on OpenQA.
Test run can be altered using environment variables:
DEFAULT_LVM_POOL- LVM thin pool to use for tests, in
QUBES_TEST_PCIDEV- PCI device to be used in PCI passthrough tests (for example sound card)
QUBES_TEST_TEMPLATES- space separated list of templates to run tests on; if not set, all installed templates are tested
QUBES_TEST_LOAD_ALL- load all tests (including tests for all templates) when relevant test modules are imported; this needs to be set for test runners not supporting load_tests protocol
Adding a new test to core-admin
You’ll also need to add your test at the bottom of the
__init__.py file, in the method
def load_tests, in the for loop with
Again, given the hypothetical
for modname in ( 'qubes.tests.basic', 'qubes.tests.dom0_update', 'qubes.tests.network', 'qubes.tests.vm_qrexec_gui', 'qubes.tests.backup', 'qubes.tests.backupcompatibility', 'qubes.tests.regressions', 'qubes.tests.example', # This is our newly added test ):
Testing PyQt applications
When testing (Py)QT application, it’s useful to create separate QApplication object for each test. But QT framework does not allow to have multiple QApplication objects in the same process at the same time. This means it’s critical to reliably cleanup previous instance before creating the new one. This turns out to be non-trivial task, especially if any test uses event loop. Failure to perform proper cleanup in many cases results in SEGV. Below you can find steps for the proper cleanup:
import asyncio import quamash import unittest import gc class SomeTestCase(unittest.TestCase): def setUp(self): [...] # force "cleanlooks" style, the default one on Xfce (GtkStyle) use # static variable internally and caches pointers to later destroyed # objects (result: SEGV) self.qtapp = QtGui.QApplication(["test", "-style", "cleanlooks"]) # construct event loop even if this particular test doesn't use it, # otherwise events with qtapp references will be queued there anyway and the # first test that actually use event loop will try to dereference (already # destroyed) objects, resulting in SEGV self.loop = quamash.QEventLoop(self.qtapp) def tearDown(self): [...] # process any pending events before destroying the object self.qtapp.processEvents() # queue destroying the QApplication object, do that for any other QT # related objects here too self.qtapp.deleteLater() # process any pending events (other than just queued destroy), just in case self.qtapp.processEvents() # execute main loop, which will process all events, _including just queued destroy_ self.loop.run_until_complete(asyncio.sleep(0)) # at this point it QT objects are destroyed, cleanup all remaining references; # del other QT object here too self.loop.close() del self.qtapp del self.loop gc.collect()
Installation Tests with openQA
Manually testing the installation of Qubes OS is a time-consuming process. We use openQA to automate this process. It works by installing Qubes in KVM and interacting with it as a user would, including simulating mouse clicks and keyboard presses. Then, it checks the output to see whether various tests were passed, e.g., by comparing the virtual screen output to screenshots of a successful installation.
Using openQA to automatically test the Qubes installation process works as of Qubes 4.0-rc4 on 2018-01-26, provided that the versions of KVM and QEMU are new enough and the hardware has VT-x and EPT. KVM also supports nested virtualization, so HVM should theoretically work. In practice, however, either Xen or QEMU crashes when this is attempted. Nonetheless, PV works well, which is sufficient for automated installation testing.
Thanks to an anonymous donor, our openQA system is hosted in a datacenter on hardware that meets these requirements.