# -*- coding: utf-8 -*- # Import python libs from __future__ import absolute_import, print_function, unicode_literals import errno # Import Salt libs import salt.utils.systemd as _systemd from salt.exceptions import SaltInvocationError from tests.support.mock import Mock, patch # Import Salt Testing libs from tests.support.unit import TestCase def _booted_effect(path): return path == "/run/systemd/system" def _not_booted_effect(path): if path == "/run/systemd/system": raise OSError(errno.ENOENT, "No such file or directory", path) class SystemdTestCase(TestCase): """ Tests the functions in salt.utils.systemd """ def test_booted(self): """ Test that salt.utils.systemd.booted() returns True when minion is systemd-booted. """ # Ensure that os.stat returns True. os.stat doesn't return a bool # normally, but the code is doing a simple truth check on the return # data, so it is sufficient enough to mock it as True for these tests. with patch("os.stat", side_effect=_booted_effect): # Test without context dict passed self.assertTrue(_systemd.booted()) # Test that context key is set when context dict is passed context = {} self.assertTrue(_systemd.booted(context)) self.assertEqual(context, {"salt.utils.systemd.booted": True}) def test_not_booted(self): """ Test that salt.utils.systemd.booted() returns False when minion is not systemd-booted. """ # Ensure that os.stat raises an exception even if test is being run on # a systemd-booted host. with patch("os.stat", side_effect=_not_booted_effect): # Test without context dict passed self.assertFalse(_systemd.booted()) # Test that context key is set when context dict is passed context = {} self.assertFalse(_systemd.booted(context)) self.assertEqual(context, {"salt.utils.systemd.booted": False}) def test_booted_return_from_context(self): """ Test that the context data is returned when present. To ensure we're getting data from the context dict, we use a non-boolean value to differentiate it from the True/False return this function normally produces. """ context = {"salt.utils.systemd.booted": "foo"} self.assertEqual(_systemd.booted(context), "foo") def test_booted_invalid_context(self): """ Test with invalid context data. The context value must be a dict, so this should raise a SaltInvocationError. """ # Test with invalid context data with self.assertRaises(SaltInvocationError): _systemd.booted(99999) def test_version(self): """ Test that salt.utils.systemd.booted() returns True when minion is systemd-booted. """ with patch("subprocess.Popen") as popen_mock: _version = 231 output = "systemd {0}\n-SYSVINIT".format(_version) popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: (output, None), pid=lambda: 12345, retcode=0, ) # Test without context dict passed self.assertEqual(_systemd.version(), _version) # Test that context key is set when context dict is passed context = {} self.assertTrue(_systemd.version(context)) self.assertEqual(context, {"salt.utils.systemd.version": _version}) def test_version_generated_from_git_describe(self): """ Test with version string matching versions generated by git describe in systemd. This feature is used in systemd>=241. """ with patch("subprocess.Popen") as popen_mock: _version = 241 output = "systemd {0} ({0}.0-0-dist)\n-SYSVINIT".format(_version) popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: (output, None), pid=lambda: 12345, retcode=0, ) # Test without context dict passed self.assertEqual(_systemd.version(), _version) # Test that context key is set when context dict is passed context = {} self.assertTrue(_systemd.version(context)) self.assertEqual(context, {"salt.utils.systemd.version": _version}) def test_version_return_from_context(self): """ Test that the context data is returned when present. To ensure we're getting data from the context dict, we use a non-integer value to differentiate it from the integer return this function normally produces. """ context = {"salt.utils.systemd.version": "foo"} self.assertEqual(_systemd.version(context), "foo") def test_version_invalid_context(self): """ Test with invalid context data. The context value must be a dict, so this should raise a SaltInvocationError. """ # Test with invalid context data with self.assertRaises(SaltInvocationError): _systemd.version(99999) def test_version_parse_problem(self): """ Test with invalid context data. The context value must be a dict, so this should raise a SaltInvocationError. """ with patch("subprocess.Popen") as popen_mock: popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: ("invalid", None), pid=lambda: 12345, retcode=0, ) # Test without context dict passed self.assertIsNone(_systemd.version()) # Test that context key is set when context dict is passed. A failure # to parse the systemctl output should not set a context key, so it # should not be present in the context dict. context = {} self.assertIsNone(_systemd.version(context)) self.assertEqual(context, {}) def test_has_scope_systemd204(self): """ Scopes are available in systemd>=205. Make sure that this function returns the expected boolean. We do three separate unit tests for versions 204 through 206 because mock doesn't like us altering the return_value in a loop. """ with patch("subprocess.Popen") as popen_mock: _expected = False _version = 204 _output = "systemd {0}\n-SYSVINIT".format(_version) popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: (_output, None), pid=lambda: 12345, retcode=0, ) # Ensure that os.stat returns True. os.stat doesn't return a bool # normally, but the code is doing a simple truth check on the # return data, so it is sufficient enough to mock it as True for # these tests. with patch("os.stat", side_effect=_booted_effect): # Test without context dict passed self.assertEqual(_systemd.has_scope(), _expected) # Test that context key is set when context dict is passed context = {} self.assertEqual(_systemd.has_scope(context), _expected) self.assertEqual( context, { "salt.utils.systemd.booted": True, "salt.utils.systemd.version": _version, }, ) def test_has_scope_systemd205(self): """ Scopes are available in systemd>=205. Make sure that this function returns the expected boolean. We do three separate unit tests for versions 204 through 206 because mock doesn't like us altering the return_value in a loop. """ with patch("subprocess.Popen") as popen_mock: _expected = True _version = 205 _output = "systemd {0}\n-SYSVINIT".format(_version) popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: (_output, None), pid=lambda: 12345, retcode=0, ) # Ensure that os.stat returns True. os.stat doesn't return a bool # normally, but the code is doing a simple truth check on the # return data, so it is sufficient enough to mock it as True for # these tests. with patch("os.stat", side_effect=_booted_effect): # Test without context dict passed self.assertEqual(_systemd.has_scope(), _expected) # Test that context key is set when context dict is passed context = {} self.assertEqual(_systemd.has_scope(context), _expected) self.assertEqual( context, { "salt.utils.systemd.booted": True, "salt.utils.systemd.version": _version, }, ) def test_has_scope_systemd206(self): """ Scopes are available in systemd>=205. Make sure that this function returns the expected boolean. We do three separate unit tests for versions 204 through 206 because mock doesn't like us altering the return_value in a loop. """ with patch("subprocess.Popen") as popen_mock: _expected = True _version = 206 _output = "systemd {0}\n-SYSVINIT".format(_version) popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: (_output, None), pid=lambda: 12345, retcode=0, ) # Ensure that os.stat returns True. os.stat doesn't return a bool # normally, but the code is doing a simple truth check on the # return data, so it is sufficient enough to mock it as True for # these tests. with patch("os.stat", side_effect=_booted_effect): # Test without context dict passed self.assertEqual(_systemd.has_scope(), _expected) # Test that context key is set when context dict is passed context = {} self.assertEqual(_systemd.has_scope(context), _expected) self.assertEqual( context, { "salt.utils.systemd.booted": True, "salt.utils.systemd.version": _version, }, ) def test_has_scope_no_systemd(self): """ Test the case where the system is not systemd-booted. We should not be performing a version check in these cases as there is no need. """ with patch("os.stat", side_effect=_not_booted_effect): # Test without context dict passed self.assertFalse(_systemd.has_scope()) # Test that context key is set when context dict is passed. # Because we are not systemd-booted, there should be no key in the # context dict for the version check, as we shouldn't have # performed this check. context = {} self.assertFalse(_systemd.has_scope(context)) self.assertEqual(context, {"salt.utils.systemd.booted": False}) def test_has_scope_version_parse_problem(self): """ Test the case where the system is systemd-booted, but we failed to parse the "systemctl --version" output. """ with patch("subprocess.Popen") as popen_mock: popen_mock.return_value = Mock( communicate=lambda *args, **kwargs: ("invalid", None), pid=lambda: 12345, retcode=0, ) with patch("os.stat", side_effect=_booted_effect): # Test without context dict passed self.assertFalse(_systemd.has_scope()) # Test that context key is set when context dict is passed. A # failure to parse the systemctl output should not set a context # key, so it should not be present in the context dict. context = {} self.assertFalse(_systemd.has_scope(context)) self.assertEqual(context, {"salt.utils.systemd.booted": True}) def test_has_scope_invalid_context(self): """ Test with invalid context data. The context value must be a dict, so this should raise a SaltInvocationError. """ # Test with invalid context data with self.assertRaises(SaltInvocationError): _systemd.has_scope(99999)