1# Autotest Best Practices 2When the Chrome OS team started using autotest, we tried our best to figure out 3how to fit our code and our tests into the upstream style with little guidance 4and poor documentation. This went poorly. With the benefit of hindsight, 5we’re going to lay out some best-practices that we’d like to enforce going 6forward. In many cases, there is legacy code that contradicts this style; we 7should go through and refactor that code to fit these guidelines as time 8allows. 9 10## Upstream Documentation 11 12There is a sizeable volume of general Autotest documentation available on 13github: 14https://github.com/autotest/autotest/wiki 15 16## Coding style 17 18Basically PEP-8. See [docs/coding-style.md](docs/coding-style.md) 19 20## Where should my code live? 21 22| Type of Code | Relative Path | 23|---------------------------|-------------------------| 24| client-side tests | client/site_tests/ | 25| server-side tests | server/site_tests | 26| common library code | client/common_lib/cros/ | 27| server-only library code | server/cros | 28 29 30## Writing tests 31 32An autotest is really defined by its control file. A control file contains 33important metadata about the test (name, author, description, duration, what 34suite it’s in, etc) and then pulls in and executes the actual test code. This 35test code can be shared among multiple distinct test cases by parameterizing it 36and passing those parameters in from separate control files. 37 38Autotests *must*: 39 40 * Be self-contained: assume nothing about the condition of the device 41 * Be hermetic: requiring the Internet to be reachable in order for your test 42 to succeed is unacceptable. 43 * Be automatic: avoid user interaction and run-time specification of input 44 values. 45 * Be integration tests: if you can test the feature in a unit test (or a 46 chrome browser test), do so. 47 * Prefer object composition to inheritance: avoid subclassing test.test to 48 implement common functionality for multiple tests. Instead, create a class 49 that your tests can instantiate to perform common operations. This enables 50 us to write tests that use both PyAuto and Servo without dealing with 51 multiple inheritance, for example. 52 * Be deterministic: a test should not validate the timing of some operation. 53 Instead, write a test that records the timing in performance keyvals so that 54 we can track the numbers over time. 55 56Autotests *must not*: 57 58 * Put significant logic in the control file: control files are really just 59 python, so one can put arbitrary logic in there. Don’t. Run your test 60 code, perhaps with some parameters. 61 62Autotests *may*: 63 64 * Share parameterized fixtures: a test is defined by a control file. Control 65 files import and run test code, and can pass simple parameters to the code 66 they run through a well-specified interface. 67 68Autotest has a notion of both client-side tests and server-side tests. Code in 69a client-side test runs only on the device under test (DUT), and as such isn’t 70capable of maintaining state across reboots or handling a failed suspend/resume 71and the like. If possible, an autotest should be written as a client-side 72test. A ‘server’ test runs on the autotest server, but gets assigned a DUT 73just like a client-side test. It can use various autotest primitives (and 74library code written by the CrOS team) to manipulate that device. Most, if not 75all, tests that use Servo or remote power management should be server-side 76tests, as an example. 77 78Adding a test involves putting a control file and a properly-written test 79wrapper in the right place in the source tree. There are conventions that must 80be followed, and a variety of primitives available for use. When writing any 81code, whether client-side test, server-side test, or library, have a strong 82bias towards using autotest utility code. This keeps the codebase consistent. 83 84 85## Writing a test 86 87This section explains considerations and requirements for any autotest, whether 88client or server. 89 90### Control files 91 92Upstream documentation 93Our local conventions for autotest control files deviate from the above a bit, 94but the indication about which fields are mandatory still holds. 95 96| Variable | Required | Value | 97|--------------|----------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| 98| AUTHOR | Yes | A comma-delimited string of at least one responsible engineer and a backup engineer -- or at worst a backup mailing list. i.e. AUTHOR = ‘msb, snanda’ | 99| DEPENDENCIES | No | list of tags known to the HW test lab. | 100| DOC | Yes | Long description of the test, pass/fail criteria | 101| NAME | Yes | Display name of the test. Generally this is the directory where your test lives e.g. hardware_TPMCheck. If you are using multiple run_test calls in the same control file or multiple control files with one test wrapper in the same suite, problems arise with the displaying of your test name. crosbug.com/35795. When in doubt ask. | 102| SYNC\_COUNT | No | Integer >= 1. Number of simultaneous devices needed for a test run. | 103| TIME | Yes | Test duration: 'FAST' (<1m), 'MEDIUM' (<10m), 'LONG' (<20m), 'LENGTHY' (>30m) | 104| TEST\_TYPE | Yes | Client or Server | 105| SUITE | No | A comma-delimited string of suite names that this test should be a part of. | 106 107### Choosing a Suite 108 109Currently existing suites are defined in the test\_suites/ subdirectory at the 110top level of the autotest repo. Read the docstrings there to see if your new 111test fits into one that’s already defined. 112 113When first adding a test, it should not go into the BVT suite. A test should 114only be added to the BVT after it has been running in some non-BVT suite long 115enough to establish a track record showing that the test does not fail when run 116against working software. A suite named experimental exists for tests intended 117for the BVT, and for which there is no more convenient home. 118 119### Pure python 120 121Lie, cheat and steal to keep your tests in pure python. It will be easier to 122debug failures, it will be easier to generate meaningful error output, it will 123be simpler to get your tests installed and run, and it will be simpler for the 124lab team to build tools that allow you to quickly iterate. 125 126Shelling out to existing command-line tools is done fairly often, and isn’t a 127terrible thing. The test author can wind up having to do a lot of output 128parsing, which is often brittle, but this can be a decent tradeoff in lieu of 129having to reimplement large pieces of functionality in python. 130 131Note that you will need to be sure that any commands you use are installed on 132the host. For a client-side test, “the host” means “the DUT”. For a 133server-side test, “the host” typically means “the system running autoserv”; 134however, if you use SiteHost.run(), the command will run on the DUT. On the 135server, your tests will have access to all tools common to both a typical CrOS 136chroot environment and standard Goobuntu. 137 138If you want to use a tool on the DUT, it may be appropriate to include it as a 139dependency of the chromeos-base/chromeos-test package. This ensures that the 140tool is pre-installed on every test image for every device, and will always be 141available for use. Otherwise, the tool must be installed as an autotest “dep”. 142 143_Never install your own shell scripts and call them._ Anything you can do in 144shell, you can do in python. 145 146### Reporting failures 147 148Autotest supports several kinds of failure statuses: 149 150| Status | Exception | Reason | 151|----------|-------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| 152| WARN | error.TestWarn | error.TestWarn should be used when side effects to the test running are encountered but are not directly related to the test running. For example, if you are testing Wifi and powerd crashes. *Currently* there are not any clear usecases for this and error.TestWarn should be generally avoided until further notice. | 153| TEST\_NA | error.TestNAError | This test does not apply in the current environment. | 154| ERROR | error.TestError | The test was unable to validate the desired behavior. | 155| FAIL | error.TestFail | The test determined the desired behavior failed to occur. | 156 157 158### Considerations when writing client-side tests 159 160All client-side tests authored at Google must live in the client/site\_tests sub-directory of the autotest source tree. 161 162###Compiling and executing binaries 163 164It is possible to compile source that’s included with your test and use the 165products at test runtime. The build infrastructure will compile this code for 166the appropriate target architecture and package it up along with the rest of 167your test’s resources, but this increases your development iteration time as 168you need to actually re-build and re-package your test to deploy it to the 169device. While we hope to improve tooling support for this use case in the 170future, avoiding this issue is the ideal. 171 172If you can’t avoid this, here’s how to get your code compiled and installed as 173a part of your test: 1741. Create a src/ directory next to your control file. 1752. Put your source, including its Makefile, in src/ 1763. define a method in your test class called “setup(self)” that takes no arguments. 1774. setup(self) should perform all tasks necessary to build your tool. There are some helpful utility functions in client/common_lib/base_utils.py. Trivial example: 178 179``` 180 def setup(self): 181 os.chdir(self.srcdir) 182 utils.make('OUT_DIR=.') 183``` 184 185### Reusing code (“fixtures”) 186 187Any autotest is, essentially, a single usage of a re-usable test fixture. This 188is because run\_once() in your test wrapper can take any arguments you want. As 189such, multiple control files can re-use the same wrapper -- and should, where 190it makes sense. 191 192### Considerations when writing server-side tests 193 194All server-side tests authored at Google must live in the server/site\_tests 195sub-directory of the autotest source tree. 196 197It should be even easier to keep the server-side of a test in pure python, as 198you should simply be driving the DUT and verifying state. 199 200### When/why to write a server-side test 201 202Server-side tests are appropriate when some operation in the test can't be 203executed on the DUT. The prototypical example is rebooting the DUT. Other 204examples include tests that manipulate the network around the DUT (e.g. WiFi 205tests), tests that power off the DUT, and tests that rely on a Servo attached 206to the DUT. 207 208One simple criterion for whether to write a server-side test is this: Is the 209DUT an object that the test must manipulate? If the answer is “yes”, then a 210server-side test makes sense. 211 212### Control files for server-side tests 213 214Server-side tests commonly operate on the DUT as an object. Autotest 215represents the DUT with an instance of class Host; the instance is constructed 216and passed to the test from the control file. Creating the host object in the 217control file can be done using certain definitions present in the global 218environment of every control file: 219 220 * Function hosts.create\_host() will create a host object from a string with 221 the name of the host (an IP address as a string is also acceptable). 222 * Variable machines is a list of the host names available to the test. 223 224Below is a sample fragment for a control file that runs a simple server side test in parallel on all the hosts specified for the test. The fragment is a complete control file, except for the missing boilerplate comments and documentation definitions required in all control files. 225 226``` 227def run(machine): 228 host = hosts.create_host(machine) 229 job.run_test("platform_ServerTest", host=host) 230 231parallel_simple(run, machines) 232``` 233 234Note: The sample above relies on a common convention that the run\_once() 235method of a server-side test defines an argument named host with a default 236value, e.g. 237 238``` 239def run_once(self, host=None): 240 # … test code goes here. 241``` 242 243### Operations on Host objects 244 245A Host object supports various methods to operate on a DUT. Below is a short list of important methods supported by instances of Host: 246 247 * run(command) - run a shell command on the host 248 * reboot() - reboot the host, and wait for it to be back on the network 249 * wait_up() - wait for the host to be active on the network 250 * wait_down() - wait until the host is no longer on the network, or until it is known to have rebooted. 251 252More details, including a longer list of available methods, and more about how 253they work can be found in the Autotest documentation for autoserv and Autotest 254documentation for Host. 255 256### Servo-based tests 257 258For server-side tests that use a servo-attached DUT, the host object has a 259servo attribute. If Autotest determines that the DUT has a Servo attached, the 260servo attribute will be a valid instance of a Servo client object; otherwise 261the attribute will be None. 262 263For a DUT in the lab, Autotest will automatically determine whether there is a 264servo available; however, if a test requires Servo, its control file must have 265additional code to guarantee a properly initialized servo object on the host. 266 267Below is a code snippet outlining the requirements; portions of the control file have been omitted for brevity: 268 269``` 270# ... Standard boilerplate variable assignments... 271DEPENDENCIES = "servo" 272# ... more standard boilerplate... 273 274args_dict = utils.args_to_dict(args) 275servo_args = hosts.SiteHost.get_servo_arguments(args_dict) 276 277def run(machine): 278 host = hosts.create_host(machine, servo_args=servo_args) 279 job.run_test("platform_SampleServoTest", host=host) 280 281parallel_simple(run, machines) 282``` 283 284The `DEPENDENCIES` setting guarantees that if the test is scheduled in the lab, 285it will be assigned to a DUT that has a servo. 286 287The setting of `servo_args` guarantees two distinct things: First, it forces 288checks that will make sure that the Servo is functioning properly; this 289guarantees that the host's `servo` attribute will not be None. Second, the code 290allows you to pass necessary servo specific command-line arguments to 291`test_that`. 292 293If the test control file follows the formula above, the test can be reliably called in a variety of ways: 294 * When used for hosts in the lab, the host’s servo object will use the servo attached to the host, and the test can assume that the servo object is not None. 295 * If you start servod manually on your desktop using the default port, you can use test_that without any special options. 296 * If you need to specify a non-default host or port number (e.g. because servod is remote, or because you have more than one servo board), you can specify them with commands like these: 297 298``` 299test_that --args=”servo_host=...” … 300test_that --args=”servo_port=...” … 301test_that --args=”servo_host=... servo_port=...” ... 302``` 303 304### Calling client-side tests from a server-side test 305 306Commonly, server-side tests need to do more on the DUT than simply run short 307shell commands. In those cases, a client-side test should be written and 308invoked from the server-side test. In particular, a client side test allows 309the client side code to be written in Python that uses standard Autotest 310infrastructure, such as various utility modules or the logging infrastructure. 311 312Below is a short snippet showing the standard form for calling a client-side 313test from server-side code: 314 315``` 316from autotest_lib.server import autotest 317 318 # ... inside some function, e.g. in run_once() 319 client_at = autotest.Autotest(host) 320 client_at.run_test("platform_ClientTest") 321``` 322 323### Writing library code 324 325There is a large quantity of Chromium OS specific code in the autotest 326codebase. Much of this exists to provide re-usable modules that enable tests 327to talk to system services. The guidelines from above apply here as well. 328This code should be as pure python as possible, though it is reasonable to 329shell out to command line tools from time to time. In some cases we’ve done 330this where we could (now) use the service’s DBus APIs directly. If you’re 331adding code to allow tests to communicate with your service, it is strongly 332recommended that you use DBus where possible, instead of munging config files 333directly or using command-line tools. 334 335Currently, our library code lives in a concerning variety of places in the 336autotest tree. This is due to a poor initial understanding of how to do 337things, and new code should follow the following conventions instead: 338 339 * Used only in server-side tests: server/cros 340 * Used in both server- and client-side tests, or only client: 341 client/common\_lib/cros 342 343### Adding test deps 344 345This does not refer to the optional `DEPENDENCIES` field in test control files. 346Rather, this section discusses how and when to use code/data/tools that are not 347pre-installed on test images, and should (or can) not be included right in with 348the test source. 349 350Unfortunately, there is no hard-and-fast rule here. Generally, if this is some 351small tool or blob of data you need for a single test, you should include it as 352discussed above in Writing client-side tests. If you’re writing the tool, and 353it has use for developers as well as in one or more tests that you’re writing, 354then make it a first-class CrOS project. Write an ebuild, write unit tests, 355and then add it to the test image by default. This can be done by RDEPENDing 356on your new test package from the chromeos-test ebuild. 357 358If your code/data falls in the middle (useful to several tests, not to devs), 359and/or is large (hundreds of megabytes as opposed to tens) then using an 360autotest ‘dep’ may be the right choice. Conceptually, an autotest test dep is 361simply another kind of archive that the autotest infrastructure knows how to 362fetch and unpack. There are two components to including a dependency from an 363autotest test -- setup during build time and installing it on your DUT when 364running a test. The setup phase must be run from your tests setup() method 365like so: 366 367``` 368def setup(self): 369 self.job.setup_dep([‘mydep’]) 370 logging.debug(‘mydep is at %s’ % (os.path.join(self.autodir, 371 ‘deps/mydep’)) 372``` 373 374The above gets run when you “build” the test. 375 376The other half of this equation is actually installing the dependency so you 377can use it while running a test. To do this, add the following to either your 378run\_once or initialize methods: 379 380``` 381 dep = dep_name 382 dep_dir = os.path.join(self.autodir, 'deps', dep=dep) 383 self.job.install_pkg(dep, 'dep', dep_dir) 384``` 385 386 387You can now reference the content of your dep using dep_dir. 388 389Now that you know how to include a dep, the next question is how to write one. 390Before you read further, you should check out client/deps/\* for many examples 391of deps in our autotest tree. 392 393### Create a dep from a third-party package 394 395There are many examples of how to do this in the client/deps directory already. 396The key component is to check in a tarball of the version of the dependency 397you’d like to include under client/deps/your\_dep. 398 399All deps require a control file and an actual python module by the same name. 400They will also need a copy of common.py to import utils.update\_version. Both 401the control and common are straightforward, the python module does all the 402magic. 403 404The deps python module follows a standard convention: a setup function and a 405call to utils.update\_version. update\_version is used instead of directly 406calling setup as it maintains additional versioning logic ensuring setup is 407only done 1x per dep. The following is its method signature: 408 409``` 410def update_version(srcdir, preserve_srcdir, new_version, install, 411 *args, **dargs) 412``` 413 414 415Notably, install should be a pointer to your setup function and `*args` should 416be filled in with params to said setup function. 417 418If you are using a tarball, your setup function should look something like: 419 420``` 421def setup(tarball, my_dir) 422 utils.extract_tarball_to_dir(tarball, my_dir) 423 os.chdir(my_dir) 424 utils.make() # this assumes your tarball has a Makefile. 425``` 426 427And you would invoke this with: 428 429``` 430utils.update_version(os.getcwd(), True, version, setup, tarball_path, 431 os.getcwd()) 432``` 433 434 435Note: The developer needs to call this because def setup is a function they are 436defining that can take any number of arguments or install the dep in any way 437they see fit. The above example uses tarballs but some are distributed as 438straight source under the src dir so their setup function only takes a top 439level path. We could avoid this by forcing a convention but that would be 440artificially constraining the deps mechanism. 441 442Once you’ve created the dep, you will also have to add the dep to the 443autotest-deps package in chromiumos-overlay/chromeos-base/autotest-deps, 444‘cros\_workon start’ it, and re-emerge it. 445 446### Create a dep from other chrome-os packages 447 448One can also create autotest deps from code that lives in other CrOS packages, 449or from build products generated by other packages. This is similar as above 450but you can reference code using the `CHROMEOS_ROOT` env var that points to the 451root of the CrOS source checkout, or the SYSROOT env var (which points to 452/build/<board>) to refer to build products. Again, read the above. Here’s an 453example of the former with the files I want in 454chromeos\_tree/chromite/my\_dep/\* where this will be the python code in 455autotest/files/client/deps/my\_dep/my\_dep.py module. 456 457``` 458import common, os, shutil 459from autotest_lib.client.bin import utils 460 461version = 1 462 463def setup(setup_dir): 464 my_dep_dir = os.path.join(os.environ['CHROMEOS_ROOT'], 'chromite', 465 'buildbot') 466 shutil.copytree(my_dep_dir, setup_dir) 467 468 469work_dir = os.path.join(os.getcwd(), 'src') 470utils.update_version(os.getcwd(), True, version, setup, work_dir) 471``` 472