Autotest Documentation For Enterprise

To provide all the information needed about the current state of Enterprise autotest automation. Current coverage, location of tests, how to execute the tests, what machine to run the tests on, test breakdown, etc.

Current coverage

Calculating coverage could be tricky as there are many different ways it could be done. We were using two ways to do it:

  • By policy:
    • Look at this recently updated spreadsheet: There are 265 policies available for ChromeOS via C/D Panel. We have 96 policies automated, 75 of those are in C/D Panel. So that’s 75/264 = %28 coverage + 21 more tests covering various other policies.
  • By section:
    • Refer to this recently updated spreadsheet in which we list out current coverage.

Test Location

  • Tests that automate user policies are located here.
  • Tests that automate device policies are located here.
  • Most of Enterprise tests start with policy_ but there are some that begin with enterprise_.

Test Results

  • The best way to view test results is by using stainless:
  • Go to https://stainless.corp.google.com/
  • Click on Test History Matrix
  • In the Test dropdown, select “policy_*”
  • Hit Search and you should see the results like so: Results

Running a test

A test can be executed using this command from chroot:

test_that --board=$BOARD_NAME $IP_ADDRESS FULL_TEST_NAME*

Example:

/trunk/src/scripts $ test_that --board=hana 100.107.106.138
policy_DeviceServer.AllowBluetooth_true

--board - should be the board that you have setup locally. You only need to setup the board ones and you shouldn’t have to touch it again for a long time. The board that you setup on your workstation doesn’t have to match the DUT(device under test) board that you’re executing the test on. To set up the board please follow instructions here. You will also need to run the build_packages command.

IP_ADDRESS - IP of the DUT. If you have a device locally, it needs to be plugged into the test network and not corp network. You can also use a device in the lab. To reserve a device from the lab please follow these steps:

  • Setup skylab using go/skylab-tools-guide (Advanced users: Manual installation)
  • “Lease” a dut go/skylab-dut-locking
  • Grab the host name, for example: chromeos15-row3-rack13-host2. Do not include the prefix (e.g. “crossk”)
  • Use this as the IP: chromeos15-row3-rack13-host2**.cros**.

Full test name - test name can be grabbed from the control file. Example.

You can check other options for test_that by running: test_that --help.

Setting up a local DUT

To run a test on a local DUT you need to make sure the DUT has been properly setup with a test build. You can use this helpful tool. Execute from here Run this command to put the build on a USB stick:

*./crosdl.py -c dev -t -b 12503.0.0 -p sarien --to_stick /dev/sda*

Or this command to update the DUT directly(flaky):

*./crosdl.py -c dev -t -b 12105.54.0 -p sarien --to_ip $IP_ADDRESS*

Note: The DUT must be reachable via SSH for this to work.

To find out the right build number, please use goldeneye and search for the right build for your board.

Test Breakdown

See the Autotest Best Practices for general autotest information. This section will provide details on how Enterprise autotests are written. Each test will require the following:

  • A control file
  • Control files for each test configuration
  • A .py defining the test, which inherits test.test

Control files

Control files are used as the entry point to a test. A typical dir for a user policy (client) test will consist of control file(s) and, along with .py test file(s). A control file will contain basic description of the test as well as options such as these:

	AUTHOR = 'name'
	NAME = 'full_test_name'
	ATTRIBUTES = 'suite:ent-nightly, suite:policy'
	TIME = 'SHORT'
	TEST_CATEGORY = 'General'
	TEST_CLASS = 'enterprise'
	TEST_TYPE = 'client'

On a user policy (client) test, there will be a base control file, plus an additional file for each test configuration. Example In this example there is the “base” control file, with no args specified, which is simply named “control”. Additionally there is a control file for each configuration of the test (.allow, .disallow, .not_set). The args to be passed to the test (.py) are specified in the final line of each of those control files. Example:

job.run_test('policy_AllowDinosaurEasterEgg',
             case=True)

Test file

Example of a basic test. The class name of the test, policy_ShowHomeButton has to match the name of the .py file, and should ideally match the directory name as well.

run_once - The function that gets called first. Parameters from the control passed into this function.

setup_case - sets up DMS, logs in, verifies policies values and various other login arguments. Defined: enterprise_policy_base. Explained in detail below.

start_ui_root - needed if you’re planning on interacting with UI objects during your test. Defined:ui_utils. This CL describes what ui_utils is based off and the usefulness of it.

check_home_button - Function that verifies the presence of the Home button in this test. Depending on the policy setting, the test is using ui.item_present to verify the status of the Home button.

Every enterprise test will require a run_once function and will most likely require setup_case. You will need to pass in a dictionary with the policy name and value into setup_case.

Useful utility

This utils.py file which contains many useful functions that you’ll come across in tests.

Some examples:

  • poll_for_condition - keeps checking for condition to be true until a timeout is reached at which point an error is raised.
  • run - runs a shell command on the DUT.

Difference between device policy test and user policy test

To run test device policies the DUT will need to be fully enrolled, starting with a cleared TPM (thus a reboot). Client tests do not support rebooting the device before/during/after a test.

In order to support clearing the TPM & rebooting, all device policies must be written as a “server” test. Server tests (for Enterprise) will need a “server” control & test, in addition to having a client control file and a .py test file. The server test will do any server operations (reboot, servo control, wifi cell control, etc)

Below is an example of testing a device Example of the server control file. This will run the server test policy_DeviceServer and pass the parameters specified. The server test will clear the tpm, create an autotest client of the DUT, then run the autotest specified in the control file policy_DeviceAllowBluetooth.

Note The parameterization control files are all of the server control files. The Client side control file is only a pass through for the parameters from the control file, and does not set any new behavior.

Debugging an autotest

Unfortunately there‘s no good debugging tool in autotest and you can’t use pdb so you‘re left with using time.sleep and logging. With time.sleep you can pause the test and see what’s going on in the actual device. When using logging you can run ‘logging.info(“what you want to log”)’ and then when the test is done running you can check the results here: /tmp/test_that_latest/results-1-TESTNAME/TESTNAME/debug/TESTNAME.INFO

If a test is failing remotely, on stainless, you can view the logs there by clicking on the Logs link. You can also see the screenshot of the screen when a test errored/failed.

Using Servo board with Autotests

Some tests require the use of the Servo Board. If you want to get ahold of a servo board you need to reach out to crosdistros@ and request one. You can either get a Servo type A or Servo type C, in case your test involves controlling the power to the DUT.

Setting up the servo, hopefully you‘ll find this screenshot useful. You can see that two cables on the left go to the DUT and the cable on the right goes into the host machine. If you’re going to be feeding the power to the DUT you will also need to connect a Type-C charger to the Servo by plugging it into the slot marked “Dut Power”. Note: if you grabbed the micro usb -> USB A cables in the tech stop make sure that the light on the switch glows orange and not green. If it's green the tests will not work.

Starting the servo, from chroot run: “sudo servo_updater” make sure everything is up to date. Then run “sudo servod -b BOARD_NAME” BOARD_NAME being the board you have built on your server. While this is running, in another terminal tab you can now execute dut-control commands such as “dut-control servo_v4_role:scr”.

With the servod running you can now execute local tests using the servo board. Example test using servo.

Enterprise Autotest Infra

This section will focus on a basic explination of the Enterprise base class used for autotest, along with commonly used calls, APIs, etc.

Base class overview:

The enterprise base class currently supports the following:

  • Enrolling with a fake account & DMS through the full OOBE flow. Commonly used for device policy testing)
  • Kiosk enrollment with fake account
  • Enrolling for user policies (not requiring OOBE flow).
  • Enterprise ARC tests
  • Logging in with a real account/DMS
  • Enrolling with a real account- currently broken see http://crbug.com/1019320
  • Configuring User/Device/Extension policies with a fake DMS
  • Obtaining policies through an API
  • Verifying policies
  • UI interaction

In addition to the features above, the base class will setup chrome for testing. This includes passing in username/password, browser flags, ARC settings, etc.

Policy Management

Policy Managing with a fake DMS is mostly handled via the policy_manager.

The Enterprise base class uses the policy manager to configure policies, set the policies with the fake DMS server, obtain policies from a DUT, and verify they are properly set (ie match the configured). In addition the policy manager handles features such as adding/updating/removing policies once after the initial setup, and make complex testing, such as extension of obfuscated policies easier to test.

If a test is to fail with “Policy <POLICY_NAME> value was not set correctly.”, the verification within the policy_manager is failing. This means the policy that was configured via the policy_manager does not match the value obtained from the DUT.

When using the fake DMS (see enterprise_fake_dmserverand policy_testserver, policies are provided to the fDMS via a json blob which is created by the policy_manager.

Policies from the DUT are obtained via an autotestprivate API, called via the enterprise_policy_utils get_all_policies and policies are refreshed (ie force a refetch from the DMS) via refresh_policies.

Enrollment and Kiosk Mode

Enterprise autotest uses the autotest enrollment to support device enrollment.

This class has the ability to enroll both real and fake accounts, including walking through the enrollment OOBE flow. The actual interaction with the UI/APIs for login is acomplished by calling telemetry.

Additionally Kiosk mode is also supported.

Chrome

Tests interact with chrome (ie launch, define plugins, ARC settings, etc) via chrome.py. chrome.py is built upon telemetry for browser interactions. The base class will handle chrome interaction for you, however there are specific examples such as the enrollment retainment test, that will interact with chrome.py directly.

Common Issues and possible solutions

  • Historically there have been issues with DUT enrollment via APIs. As of R80-x, this should be resolved. Typically enrollment issues have an error message along the lines of: test did not pass (reason: Unhandled TimeoutException: Timed out while waiting 60s for _EnterpriseWebviewVisible.). If this error is seen, it is typically related to something during the OOBE flow, when waiting for the enterprise enrollment screen.
  • Some of the Enterprise Autotests use UI interaction/reading for the tests. These UI elements change somewhat often, and will occasionally cause these tests to break. UI based errors usually have a traceback leading to ui.utils, and can often be fixed by simply update the UI element the test is looking for.
  • Errors from chrome.py can also lead to Enterprise tests failing. This package is not directly owned by Enterprise, or anyone other group, but is a shared resource. If a test fails due to this package, it is likely up to the test owner to fix, but they should be cognisant of other teams using the package.
  • inspector_backend timeouts occasionally occur (<0.5% of all tests.) The problem is traces backto a inspector backend crash/disconnect between telemetry and the DUT.This error is well outside the scope of Enterprise autotest. Rerunning the test is likely the easiest solution