diff options
Diffstat (limited to 'meta/lib/oeqa/runtime/cases/ptest.py')
-rw-r--r-- | meta/lib/oeqa/runtime/cases/ptest.py | 114 |
1 files changed, 114 insertions, 0 deletions
diff --git a/meta/lib/oeqa/runtime/cases/ptest.py b/meta/lib/oeqa/runtime/cases/ptest.py new file mode 100644 index 0000000000..3ef9022188 --- /dev/null +++ b/meta/lib/oeqa/runtime/cases/ptest.py @@ -0,0 +1,114 @@ +# +# Copyright OpenEmbedded Contributors +# +# SPDX-License-Identifier: MIT +# + +import os +import unittest +import pprint +import datetime + +from oeqa.runtime.case import OERuntimeTestCase +from oeqa.core.decorator.depends import OETestDepends +from oeqa.core.decorator.data import skipIfNotFeature +from oeqa.runtime.decorator.package import OEHasPackage +from oeqa.utils.logparser import PtestParser + +class PtestRunnerTest(OERuntimeTestCase): + + @skipIfNotFeature('ptest', 'Test requires ptest to be in DISTRO_FEATURES') + @OETestDepends(['ssh.SSHTest.test_ssh']) + @OEHasPackage(['ptest-runner']) + @unittest.expectedFailure + def test_ptestrunner_expectfail(self): + if not self.td.get('PTEST_EXPECT_FAILURE'): + self.skipTest('Cannot run ptests with @expectedFailure as ptests are required to pass') + self.do_ptestrunner() + + @skipIfNotFeature('ptest', 'Test requires ptest to be in DISTRO_FEATURES') + @OETestDepends(['ssh.SSHTest.test_ssh']) + @OEHasPackage(['ptest-runner']) + def test_ptestrunner_expectsuccess(self): + if self.td.get('PTEST_EXPECT_FAILURE'): + self.skipTest('Cannot run ptests without @expectedFailure as ptests are expected to fail') + self.do_ptestrunner() + + def do_ptestrunner(self): + status, output = self.target.run('which ptest-runner', 0) + if status != 0: + self.skipTest("No -ptest packages are installed in the image") + + test_log_dir = self.td.get('TEST_LOG_DIR', '') + # The TEST_LOG_DIR maybe NULL when testimage is added after + # testdata.json is generated. + if not test_log_dir: + test_log_dir = os.path.join(self.td.get('WORKDIR', ''), 'testimage') + # Make the test output path absolute, otherwise the output content will be + # created relative to current directory + if not os.path.isabs(test_log_dir): + test_log_dir = os.path.join(self.td.get('TOPDIR', ''), test_log_dir) + # Don't use self.td.get('DATETIME'), it's from testdata.json, not + # up-to-date, and may cause "File exists" when re-reun. + timestamp = datetime.datetime.now().strftime('%Y%m%d%H%M%S') + ptest_log_dir_link = os.path.join(test_log_dir, 'ptest_log') + ptest_log_dir = '%s.%s' % (ptest_log_dir_link, timestamp) + ptest_runner_log = os.path.join(ptest_log_dir, 'ptest-runner.log') + + libdir = self.td.get('libdir', '') + ptest_dirs = [ '/usr/lib' ] + if not libdir in ptest_dirs: + ptest_dirs.append(libdir) + status, output = self.target.run('ptest-runner -t 450 -d \"{}\"'.format(' '.join(ptest_dirs)), 0) + os.makedirs(ptest_log_dir) + with open(ptest_runner_log, 'w') as f: + f.write(output) + + # status != 0 is OK since some ptest tests may fail + self.assertTrue(status != 127, msg="Cannot execute ptest-runner!") + + if not hasattr(self.tc, "extraresults"): + self.tc.extraresults = {} + extras = self.tc.extraresults + extras['ptestresult.rawlogs'] = {'log': output} + + # Parse and save results + parser = PtestParser() + results, sections = parser.parse(ptest_runner_log) + parser.results_as_files(ptest_log_dir) + if os.path.exists(ptest_log_dir_link): + # Remove the old link to create a new one + os.remove(ptest_log_dir_link) + os.symlink(os.path.basename(ptest_log_dir), ptest_log_dir_link) + + extras['ptestresult.sections'] = sections + + trans = str.maketrans("()", "__") + for section in results: + for test in results[section]: + result = results[section][test] + testname = "ptestresult." + (section or "No-section") + "." + "_".join(test.translate(trans).split()) + extras[testname] = {'status': result} + + failed_tests = {} + + for section in sections: + if 'exitcode' in sections[section].keys(): + failed_tests[section] = sections[section]["log"] + + for section in results: + failed_testcases = [ "_".join(test.translate(trans).split()) for test in results[section] if results[section][test] == 'FAILED' ] + if failed_testcases: + failed_tests[section] = failed_testcases + + failmsg = "" + status, output = self.target.run('dmesg | grep "Killed process"', 0) + if output: + failmsg = "ERROR: Processes were killed by the OOM Killer:\n%s\n" % output + + if failed_tests: + failmsg = failmsg + "Failed ptests:\n%s" % pprint.pformat(failed_tests) + + if failmsg: + self.logger.warning("There were failing ptests.") + self.fail(failmsg) |