From Fedora Project Wiki

(updates due to Config class change)
 
Line 340: Line 340:
         self.post_results(bodhi = {'title': name})
         self.post_results(bodhi = {'title': name})
</pre>
</pre>
==== ResultsDB ====
Results are also automagically stored into the ResultsDB <i>FIXME: add URL of resultsdb frontend</i>. The only stuff, you need to take care of are some 'identification' data. For example, when your test deals with packages/builds, you want to store <code>envr</code>, maybe <code>kojitag</code> or another usefull information.
To store this information, add appropriate key-value pairs to the <code>self.detail.keyvals</code> dictionary.
# taken from rpmguard
self.detail.keyvals = {'kojitag': kojitag, 'envr': nvr}
Even though the key-value pairs are entirely up to your decision, please make sure, that you store envr(s) of the tested packages/builds/whatever-has-envr under the 'envr' key.
Also, you can store multiple values under the same key, when the value is list of strings:
self.detail.keyvals = {'envr': ['foo', 'bar', 'foobar']}
===== Tests with multiple results per testrun =====
Some tests produce more results in one testrun (e.g. Depcheck). If you want to store them separately in ResultsDB, make sure to set the <code>self.is_multiresult</code> attribute to True.
Then you can create multiple TestDetail instances (e.g. one for each tested package/update/whatever), fill them out (don't forget the test_detail.keyvals), and post them using <code>self.post_results()</code> method.
...
all_test_details = []
for build in all_builds_to_test:
    td = TestDetail(self, id = build['nvr'], arch = arch)
    all_test_details.append(td)
    ...
    td.result = "PASSED"
    td.keyvals = {'envr': build['nvr'], ...}
    ...
for td in all_test_details:
    self.post_results(td)
...
self.test_detail.keyvals = {'envr': [build['nvr'] for build in all_builds_to_test]}
...
This will create multiple 'Testrun' records in the ResultsDB, one for each TestDetail created & reported. And one 'Job' record, which encapsulates all the 'Testruns'.


== How to run AutoQA tests ==
== How to run AutoQA tests ==

Latest revision as of 15:18, 14 March 2012

Introduction

Here's some info on writing tests for AutoQA. There's four parts to a test: the test code, the test object, the Autotest control file, and the AutoQA control file. Typically they all live in a single directory, located in the tests/ dir of the autoqa source tree.

Start with a test
Before considering integrating a test into AutoQA or Autotest, create a working test. Creating a working test should not require knowledge of autotest or autoqa. This page outlines the process of integrating an existing test into AutoQA.
No support for per-package tests
AutoQA currently supports only generic Fedora tests (tests that concern packages - or other objects - in general). We have plans to support per-package tests in the future (e.g. tests for sshd, tests for gimp, etc). Until AutoQA support exists, consider merging your test into upstream's test suite, or including your test in the Fedora rpm package, and have it executed during %check or ask the package maintainer to manually execute tests and provide results during bodhi updates.

Write test code first

I'll say it again: Write the test first. The tests don't require anything from autotest or autoqa. You should have a working test before you even start thinking about AutoQA.

You can package up pre-existing tests or you can write a new test in whatever language you're comfortable with. It doesn't even need to return a meaningful exit code if you don't want it to (even though it is definitely better). You'll handle parsing the output and returning a useful result in the test object.

If you are writing a brand new test, there are some python libraries that have been developed for use in existing AutoQA tests. More information about this will be available once these libraries are packaged correctly, but they are not necessary to write your own tests. You can choose to use whatever language and libraries you want.

The test directory

Create a new directory to hold your test. The directory name will be used as the test name, and the test object name should match that. Choose a name that doesn't use spaces, dashes, or dots. Underscores are acceptable.

Drop your test code into the directory - it can be a bunch of scripts, a tarball of sources that may need compiling, whatever.

Next, from the directory autoqa/doc/, copy template files control.template, control.autoqa.template and test_class.py.template into your test directory. Rename them to control, control.autoqa and [testname].py, respectively.

The control file

The control file defines some metadata for this test and executes the test. Here's an example control file:

control file for conflicts test

DOC = """
This test runs potential_conflict from yum-utils to check for possible
file / package conflicts.
"""
MAINTAINER = "Martin Krizek <mkrizek@redhat.com>"

job.run_test('conflicts', **autoqa_args)

Required data

The following control file items are required for valid AutoQA tests:

  • DOC: A verbose description of the test - its purpose, the logs and data it will generate, and so on.
  • MAINTAINER: The contact person for this test. If it doesn't work or there are some problems, then we know who to contact.

Launching the test object

Most tests will have a line in the control file like this:

job.run_test('conflicts', **autoqa_args)

This will create a 'conflicts' test object (see below) and pass along all required variables.

autoqa_args
A dictionary, containing all the event-specific variables (e.g. kojitag for post-koji-build event). Documentation on these is to be found in events/[eventname]/README files. Some more variables may be also present, as described in the template file.

Those variables will be inserted into the control file by the autoqa test harness when it's time to schedule the test.

The control.autoqa file

The control.autoqa file allows a test to define any scheduling requirements or modify input arguments. This file will decide whether to run this test at all, on what architectures/distributions it should run, and so on. It is evaluated on the AutoQA server before the test itself is scheduled and run on AutoQA client.

All variables available in control.autoqa are documented in doc/control.autoqa.template. You can override them to customize your test's scheduling. Basically you can influence:

  • Which event the test runs for and under which conditions.
  • The type of system the test needs. This includes system architecture, operating system version and whether the system supports virtualization (see autotest labels for additional information)
  • Data passed from the event to the test object.

Here is example control.autoqa file:

# this test can be run just once and on any architecture,
# override the default set of architectures
archs = ['noarch']

# this test may be destructive, let's require a virtual machine for it
labels = ['virt']

# we want to run this test just for post-koji-build event;
# please note that 'execute' defaults to 'False' and to have
# the test scheduled, control.autoqa needs to complete with
# 'execute' set to 'True'
if event in ['post-koji-build']:
    execute = True

Similar to the control file, the control.autoqa file is a Python script, so you can execute conditional expressions, loops or virtually any other Python statements there. However, it is heavily recommended to keep this file as simple as possible and put all the logic to the test object.

Test Object

The test object is a python file that defines an object that represents your test. It handles the setup for the test (installing packages, modifying services, etc), running the test code, and sending results to Autotest (and other places).

Convention holds that the test object file - and the object itself - should have the same name as the test. For example, the conflicts test contains a file named conflicts.py, which defines a conflicts class, as follows:

from autotest_lib.client.bin import utils
from autoqa.test import AutoQATest
from autoqa.decorators import ExceptionCatcher

class conflicts(AutoQATest):
    ...

The name of the class must match the name given in the run_test() line of the control file, and test classes must be subclasses of the AutoQATest class. But don't worry too much about how this works - the test_class.py.template contains the skeleton of an appropriate test object. Just change the name of the file (and class!) to something appropriate for your test.

AutoQATest base class

This class contains the functionality common to all the tests. When you override some of its methods (like setup(), initialize() or run_once()) it is important to call the parent method first.

The most important attribute of this class is detail (an instance of TestDetail class) that is used for storing all test outcomes.

The most important methods include log() that you are advised to use for logging test output and post_results() that you can use for changing the way your test results are reported (if you're not satisfied with the default behavior).

Whenever your test crashes, the process_exception() method will automatically catch the exception and log it (more on that in the next section).

ExceptionCatcher decorator

When an unintended exception is raised during test setup (setup()), initialization (initialize()) or execution (run_once()) and ExceptionCatcher() decorator is used for these methods (the default), it calls the process_exception() method instead of simply crashing. In this way we are able to operate and submit results even of crashed tests.

When such event occurs, the test result is set to CRASHED, the exception traceback is added into the test output and the exception info is put into the test summary. Then the results are reported in a standard way (by creating log files and sending emails). Finally the original exception is re-raised.

If a different recovery procedure than process_exception() is desired, you may define the method and provide the method name as an argument to the decorator. For example, see below:

def my_exception_handler(self, exc = None):
   '''do something different'''

@ExceptionCatcher('self.my_exception_handler')
def run_once(self, **kwargs):
    ...
**kwargs parameter
Because of some nasty Autotest magic, it is required to have the **kwargs argument in the decorated function. This is because Autotest magic can not find out what is the correct subset of arguments from **autoqa_args to pass, so it passes them all - which causes error, if you don't have them all listed.

You are advised to call the original process_exception() method inside your custom handler, if applicable.

Test stages

setup()

This is an optional method of the test class. This is where you make sure that any required packages are installed, services are started, your test code is compiled, and so on. For example:

    @ExceptionCatcher()
    def setup(self):
        retval = utils.system('yum -y install httpd')
        assert retval == 0
        if utils.system('service httpd status') != 0:
            utils.system('service httpd start')

initialize()

This does any pre-test initialization that needs to happen. AutoQA tests typically uses this method to initialize various structures, set self.detail.id and similar attributes. This is an optional method.

All basic initialization is done in the AutoQATest class, so check it out, before you re-define it.

Call AutoQATest.initialize
If you re-implement the initialize function, make sure, that you call super(self.__class__, self).initialize(config) inside it, so all the required initialization is executed

run_once()

This is where the test code actually gets run. It's the only required method for your test object.

In short, this method should build the argument list, run the test binary and process the test result and output. For example, see below:

    @ExceptionCatcher()
    def run_once(self, baseurl, parents, name, **kwargs):
        super(self.__class__, self).run_once()

        cmd = './potential_conflict.py --tempcache --newest ' \
              '--repofrompath=target,%s --repoid=target' % baseurl

        out = utils.system_output(cmd, retain_output=True)
        self.log(out, printout=False)

This above example will run the command potential_conflict.py and save its output. It will raise CmdError if the command ends with non-zero exit code.

If you need to receive just the exit code of the command, use utils.system() method instead.

Additionally, if you need both the exit code and command output, use the built-in utils.run() method:

    cmd_result = utils.run(cmd, ignore_status=True, stdout_tee=utils.TEE_TO_LOGS, stderr_tee=utils.TEE_TO_LOGS)
    output = cmd_result.stdout
    retval = cmd_result.exit_status

Useful test object attributes

AutoQATest instances have the following attributes available[1]:

outputdir       eg. results/<job>/<testname.tag>
resultsdir      eg. results/<job>/<testname.tag>/results
profdir         eg. results/<job>/<testname.tag>/profiling
debugdir        eg. results/<job>/<testname.tag>/debug
bindir          eg. tests/<test>
src             eg. tests/<test>/src
tmpdir          eg. tmp/<tempname>_<testname.tag>

Test Results

The AutoQATest class provides a detail attribute to be used for storing test results. This is an instance of TestDetail class and serves as a container for everything related to the test outcome.

If your test tests several independent things at once, you can create several TestDetail objects and then submit results for all of them manually. For example upgradepath test uses that for reporting results for every proposed Bodhi update. But this is really advanced stuff, see upgradepath or depcheck tests for inspiration.

ID

Every test run should contain a string identification of the test in self.detail.id. This doesn't have to be unique, but it should be descriptive enough for the log reader to quickly understand what has been tested. For RPM-based tests this will be probably a build NVR. For Bodhi update-based tests this will be probably a Bodhi update title. For yum repository-based tests this will be a repository name or address. And so on.

Usually the ID is set in initialize() method (ASAP) because it is not changed throughout the test:

    @ExceptionCatcher()
    def initialize(self, config, nvr, **kwargs):
        super(self.__class__, self).initialize(config)
        self.detail.id = nvr

Architecture

If your test is not an architecture-independent test (aka noarch test, the default), you must set the architecture of tested items in self.detail.arch. That is usually also done in initialize() method.

Overall Result

The overall test result is stored in self.detail.result. You should set it in run_once() according to the result of your test. You can choose from these values:

  • PASSED - the test has passed, there is no problem with it
  • INFO - the test has passed, but there is some important information that a relevant person would very probably like to review
  • FAILED - the test has failed, requirements are not met
  • NEEDS_INSPECTION (default)- the test has failed, but a relevant person is needed to inspect it and possibly may waive the errors
  • ABORTED - some third party error has occurred (networking error, external script used for testing has crashed, etc) and the test could not complete because of that. Re-running this test with same input arguments could solve this problem.
  • CRASHED - the test has crashed because of a programming error somewhere in our code (test script or autoqa code). Close inspection is necessary to be able to solve this issue.

If no value is set in self.detail.result, a value of NEEDS_INSPECTION is used.

If an exception occurs, and is caught by the ExceptionCatcher decorator (i.e. you don't catch it yourself), self.detail.result is set to CRASHED.

Using ABORTED result properly

If you want to end your test with ABORTED result, simple set self.detail.result and then re-raise the original exception. self.detail.summary will be filled-in automatically (extracted from the exception message), if empty.

try:
    //download from Koji
except IOError, e: //or some other error
    self.detail.result = 'ABORTED'
    raise

If you don't have any exception to re-raise but still want to end the test, again set self.detail.result, but this time be sure to also provide an explanation in self.detail.summary and then end the test by raising autotest_lib.client.common_lib.error.TestFail. Alternatively you can provide the error explanation as an argument to the TestFail class instead of filling in self.detail.summary.

from autotest_lib.client.common_lib import error
foo = //do some stuff
if foo == None:
    self.detail.result = 'ABORTED'
    raise error.TestFail('No result returned from service bar')

Summary

The self.detail.summary should contain a few words summarizing the test output. It is then used in the log overview and in the email subject. E.g. for conflicts test it can be "69 packages with file conflicts". Or for rpmlint test it can be "3 errors, 5 warnings". Don't repeat test name, test result or test ID in here.

Output

Log any test output you want to save (to be emailed and stored in a log file on the server) by using self.log() method. Usually this is used for saving important information throughout the test. For rpmlint test this may include information which RPM packages will be downloaded and tested, and then the very output of the rpmlint command.

self.log('Build to be tested: %s' % nvr)
cmd = '...'
output = utils.system_output(cmd, retain_output=True)
self.log(output, printout=False)
...
self.log('Found %d errors in the command output' % errors, stderr=True)

The output is stored in self.detail.output variable. You can modify it if needed (for example filter out some lines you don't want to see in the log), but you're highly discouraged from adding new content to that variable directly (use log() method instead).

Know when to print
You can also use print statements in your code. But note that these lines won't be visible in the test log (but they will be in the debug log). You can use them for printing low-level details that can potentially help you find a problem later. If you want something in between (not visible in the test log, but better accessible than debug log), consider using self.log(message, output=False). That won't show up the in the test log, but it will be visible in the full log.

Highlights

If you want to emphasize a certain line or set of lines in your output, you can do it by highlighting them. That enables the reader to easily spot warnings and errors amongst hundreds of lines of output. It is usually done by:

self.log('RPM checksum invalid', highlight=True)

You can also provide a descriptive comment to the highlighted lines:

self.log('= Problems overview =\n %s' % problems, highlight='This test will not pass until all of these problems are solved.')

If you need to highlight a line that has been already logged (e.g. part of the command output), you can directly assign to self.detail.highlights variable (it is a list of tuples (line, comment)).

Additional log fields

You can provide some additional log fields to be displayed in your log. Explore self.detail.log variable (instance of PrettyLog class). Especially these attributes:

  • summary_detail - allows you to provide additional detail for your summary section
  • kojitag/bodhi_title/bodhi_id - allows you to provide more information about which Kojitag/Bodhi update this test is related to
  • custom_fields - allows you to specify any custom field for the log overview section

Extra Data

Further test-level info can be returned by using test.write_test_keyval(dict). The following example demonstrates extracting and saving the kernel version used when running a test:

extrainfo = dict()
for line in self.results.stdout:
    if line.startswith("kernel version "):
        extrainfo['kernelver'] = line.split()[3]
    ...
self.write_test_keyval(extrainfo)

In addition to test-level key/value pairs, per-iteration key/value information (e.g. performance metrics) can be recorded:

  1. self.write_attr_keyval(attr_dict) - Store test attributes (string data). Test attributes are limited to 100 characters. [2]
  2. self.write_perf_keyval(perf_dict) - Store test performance metrics (numerical data). Performance values must be floating-point numbers.
  3. self.write_iteration_keyval(attr_dict, perf_dict) - Storing both, attributes and performance data
Not displayed
This extra data are not yet displayed anywhere in the log, but you can access them in the test results directory.

Log files and scratch data

For any test execution several logs are created. The default log is <test_id>.html and it is being linked and emailed to any configured recipients. It is designated for the large audience.

There is also an full.log. It contains everything that has been logged throughout the test execution (the default log may have some lines filtered out, but they should be always present in this log).

The most detailed log is debug log placed at debug/client.DEBUG at root test results directory. Into that file autotest automatically logs everything printed on the screen and the full output of any commands you run.

If you want to store a custom file (like your own log), just save it to self.resultsdir directory. All those files will be saved at the end of the test.

If you need to store some scratch data that should be discarded on test finish, use self.tmpdir directory for that.

Posting feedback into Bodhi

After the result of a test is known, it can be sent into Bodhi. You need to manually call post_results() method at the end of our test and provide a bodhi parameter.

        # Report test result to Bodhi
        # 'name' variable contains Bodhi update title relevant for this test
        self.post_results(bodhi = {'title': name})

ResultsDB

Results are also automagically stored into the ResultsDB FIXME: add URL of resultsdb frontend. The only stuff, you need to take care of are some 'identification' data. For example, when your test deals with packages/builds, you want to store envr, maybe kojitag or another usefull information.

To store this information, add appropriate key-value pairs to the self.detail.keyvals dictionary.

# taken from rpmguard
self.detail.keyvals = {'kojitag': kojitag, 'envr': nvr}

Even though the key-value pairs are entirely up to your decision, please make sure, that you store envr(s) of the tested packages/builds/whatever-has-envr under the 'envr' key.

Also, you can store multiple values under the same key, when the value is list of strings:

self.detail.keyvals = {'envr': ['foo', 'bar', 'foobar']}
Tests with multiple results per testrun

Some tests produce more results in one testrun (e.g. Depcheck). If you want to store them separately in ResultsDB, make sure to set the self.is_multiresult attribute to True.

Then you can create multiple TestDetail instances (e.g. one for each tested package/update/whatever), fill them out (don't forget the test_detail.keyvals), and post them using self.post_results() method.

...
all_test_details = []
for build in all_builds_to_test:
    td = TestDetail(self, id = build['nvr'], arch = arch)
    all_test_details.append(td)
    ...
    td.result = "PASSED"
    td.keyvals = {'envr': build['nvr'], ...}
    ...
for td in all_test_details:
    self.post_results(td)
...
self.test_detail.keyvals = {'envr': [build['nvr'] for build in all_builds_to_test]}
...

This will create multiple 'Testrun' records in the ResultsDB, one for each TestDetail created & reported. And one 'Job' record, which encapsulates all the 'Testruns'.

How to run AutoQA tests

Install AutoQA from GIT

First of all, you'll need to checkout some version from GIT. You can either use master, or some tagged 'release'.

To checkout master branch:

git clone git://git.fedorahosted.org/autoqa.git autoqa
cd autoqa

To checkout tagged release:

git clone git://git.fedorahosted.org/autoqa.git autoqa
cd autoqa
git tag -l
# now you'll get a list of tags, at the time of writing this document, the latests tag was v0.3.5-1
git checkout -b v0.3.5-1 tags/v0.3.5-1

Add your test

The best way to add your test into the directory structure is to create a new branch, copy your test and make install autoqa.

git checkout -b my_new_awesome_test
cp -r /path/to/directory/with/your/test ./tests
make clean install
Dependencies
It's possible that make install will fail due to missing some python modules (e.g. turbogears2). In that case install them using yum.

Run your test

This is dependent on the event, your test is supposed to run under. Let's assume, that it is the post-koji-build.

/usr/share/autoqa/post-koji-build/watch-koji-builds.py --dry-run

This command will show you current koji builds e.g.

No previous run - checking builds in the past 3 hours
autoqa post-koji-build --kojitag dist-f12-updates-candidate --arch x86_64 --arch i686 espeak-1.42.04-1.fc12
autoqa post-koji-build --kojitag dist-f11-updates-candidate --arch x86_64 kdemultimedia-4.3.4-1.fc11
autoqa post-koji-build --kojitag dist-f11-updates-candidate --arch x86_64 kdeplasma-addons-4.3.4-1.fc11
autoqa post-koji-build --kojitag dist-f12-updates-candidate --arch x86_64 --arch i686 cryptopp-5.6.1-0.1.svn479.fc12
autoqa post-koji-build --kojitag dist-f12-updates-candidate --arch x86_64 drupal-6.15-1.fc12
autoqa post-koji-build --kojitag dist-f12-updates-candidate --arch x86_64 --arch i686 seamonkey-2.0.1-1.fc12
... output trimmed ...

So to run your test, just select one of the lines, and add parameters --test name_of_your_test --local, which will locally execute the test you just wrote. If you wanted to run rpmlint, for example, the command would be

autoqa post-koji-build --kojitag dist-f12-updates-candidate --arch x86_64 --arch i686 espeak-1.42.04-1.fc12 --test rpmlint --local
--local
It is important to add the --local parameter. If you won't, the test will fail to run provided that you don't have Autotest server installed and configured.

Debugging problems

If your test behaves incorrectly in certain cases, you can use python debugger even when it is run through AutoQA/Autotest framework. Add import pdb; pdb.set_trace() line where you want to stop the program and begin debugging:

    @ExceptionCatcher()
    def run_once(self, **kwargs):
        super(self.__class__, self).run_once()

        do_some_stuff()

        import pdb
        pdb.set_trace()

        do_some_other_stuff()

Then run your test with --local option. You will be given a Pdb shell on the chosen line:

16:02:08 INFO | > /usr/share/autotest/client/site_tests/rpmlint/rpmlint.py(51)run_once()
16:02:08 INFO | -> self.log('%s\n%s\n%s' % ('='*40, nvr, '='*40))
help
16:02:22 INFO | (Pdb) 
16:02:22 INFO | Documented commands (type help <topic>):
16:02:22 INFO | ========================================
16:02:22 INFO | EOF    bt         cont      enable  jump  pp       run      unt   
16:02:22 INFO | a      c          continue  exit    l     q        s        until 
16:02:22 INFO | alias  cl         d         h       list  quit     step     up    
16:02:22 INFO | args   clear      debug     help    n     r        tbreak   w     
16:02:22 INFO | b      commands   disable   ignore  next  restart  u        whatis
16:02:22 INFO | break  condition  down      j       p     return   unalias  where 
16:02:22 INFO | 
16:02:22 INFO | Miscellaneous help topics:
16:02:22 INFO | ==========================
16:02:22 INFO | exec  pdb
16:02:22 INFO | 
16:02:22 INFO | Undocumented commands:
16:02:22 INFO | ======================
16:02:22 INFO | retval  rv
16:02:22 INFO | 
list
16:02:37 INFO | (Pdb)  46  	        super(self.__class__, self).run_once()
16:02:37 INFO |  47  	
16:02:37 INFO |  48  	        import pdb
16:02:37 INFO |  49  	        pdb.set_trace()
16:02:37 INFO |  50  	        # add header
16:02:37 INFO |  51  ->	        self.log('%s\n%s\n%s' % ('='*40, nvr, '='*40))
16:02:37 INFO |  52  	
16:02:37 INFO |  53  	        # download packages
16:02:37 INFO |  54  	        pkgs = self.koji.get_nvr_rpms(nvr, self.tmpdir, debuginfo=False,
16:02:37 INFO |  55  	                                      src=True)
16:02:37 INFO |  56  	        # run rpmlint

For a 10-minute introduction how to use Pdb read Debugging in Python.

Providing documentation

Every test should have at least basic documentation written in its control file and in its test object. But if you intend to create a test for public consumption (not just for your own needs), you should provide proper and detailed documentation on these wiki pages.

All AutoQA test documentation is in Category:AutoQA tests, and the page name should be AutoQA tests/<Test>, where test name starts with uppercase letter. These pages will be linked from the log of every executed test.

See the existing documentation for inspiration how to create your own one.

References

Links

Autotest documentation