def test_cleanup_legacy_cgroups_should_disable_cgroups_when_the_daemon_was_added_to_the_legacy_cgroup_on_systemd(self, _):
# Set up a mock /var/run/waagent.pid file
daemon_pid = "42"
daemon_pid_file = os.path.join(self.tmp_dir, "waagent.pid")
fileutil.write_file(daemon_pid_file, daemon_pid + "\n")
# Set up old controller cgroups and add the daemon PID to them
CGroupsTools.create_legacy_agent_cgroup(self.cgroups_file_system_root, "cpu", daemon_pid)
CGroupsTools.create_legacy_agent_cgroup(self.cgroups_file_system_root, "memory", daemon_pid)
# Start tracking a couple of dummy cgroups
CGroupsTelemetry.track_cgroup(CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service", "cpu"))
CGroupsTelemetry.track_cgroup(CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service", "memory"))
cgroup_configurator = CGroupConfigurator.get_instance()
with patch("azurelinuxagent.common.cgroupconfigurator.add_event") as mock_add_event:
with patch("azurelinuxagent.common.cgroupapi.get_agent_pid_file_path", return_value=daemon_pid_file):
cgroup_configurator.cleanup_legacy_cgroups()
self.assertEquals(len(mock_add_event.call_args_list), 1)
_, kwargs = mock_add_event.call_args_list[0]
self.assertEquals(kwargs['op'], 'CGroupsCleanUp')
self.assertFalse(kwargs['is_success'])
self.assertEquals(
kwargs['message'],
"Failed to process legacy cgroups. Collection of resource usage data will be disabled. [CGroupsException] The daemon's PID ({0}) was already added to the legacy cgroup; this invalidates resource usage data.".format(daemon_pid))
self.assertFalse(cgroup_configurator.enabled())
self.assertEquals(len(CGroupsTelemetry._tracked), 0)
def test_cleanup_legacy_cgroups_should_disable_cgroups_when_it_fails_to_process_legacy_cgroups(
self):
# Set up a mock /var/run/waagent.pid file
daemon_pid = "42"
daemon_pid_file = os.path.join(self.tmp_dir, "waagent.pid")
fileutil.write_file(daemon_pid_file, daemon_pid + "\n")
# Set up old controller cgroups and add the daemon PID to them
CGroupsTools.create_legacy_agent_cgroup(self.cgroups_file_system_root,
"cpu", daemon_pid)
CGroupsTools.create_legacy_agent_cgroup(self.cgroups_file_system_root,
"memory", daemon_pid)
# Set up new controller cgroups and add extension handler's PID to them
CGroupsTools.create_agent_cgroup(self.cgroups_file_system_root, "cpu",
"999")
CGroupsTools.create_agent_cgroup(self.cgroups_file_system_root,
"memory", "999")
def mock_append_file(filepath, contents, **kwargs):
if re.match(r'/.*/cpu/.*/cgroup.procs', filepath):
raise OSError(errno.ENOSPC, os.strerror(errno.ENOSPC))
fileutil.append_file(filepath, contents, **kwargs)
# Start tracking a couple of dummy cgroups
CGroupsTelemetry.track_cgroup(
CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service",
"cpu"))
CGroupsTelemetry.track_cgroup(
CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service",
"memory"))
cgroup_configurator = CGroupConfigurator.get_instance()
with patch("azurelinuxagent.common.cgroupconfigurator.add_event"
) as mock_add_event:
with patch(
"azurelinuxagent.common.cgroupapi.get_agent_pid_file_path",
return_value=daemon_pid_file):
with patch(
"azurelinuxagent.common.cgroupapi.fileutil.append_file",
side_effect=mock_append_file):
cgroup_configurator.cleanup_legacy_cgroups()
self.assertEquals(len(mock_add_event.call_args_list), 1)
_, kwargs = mock_add_event.call_args_list[0]
self.assertEquals(kwargs['op'], 'CGroupsCleanUp')
self.assertFalse(kwargs['is_success'])
self.assertEquals(
kwargs['message'],
'Failed to process legacy cgroups. Collection of resource usage data will be disabled. [Errno 28] No space left on device'
)
self.assertFalse(cgroup_configurator.enabled())
self.assertEquals(len(CGroupsTelemetry._tracked), 0)
def test_disable_should_reset_tracked_cgroups(self):
configurator = CGroupConfigurator.get_instance()
# Start tracking a couple of dummy cgroups
CGroupsTelemetry.track_cgroup(CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service", "cpu"))
CGroupsTelemetry.track_cgroup(CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service", "memory"))
configurator.disable()
self.assertFalse(configurator.enabled())
self.assertEquals(len(CGroupsTelemetry._tracked), 0)
def test_disable_should_reset_tracked_cgroups(self):
# Start tracking a couple of dummy cgroups
CGroupsTelemetry.track_cgroup(
CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service",
"cpu"))
CGroupsTelemetry.track_cgroup(
CGroup("dummy", "/sys/fs/cgroup/memory/system.slice/dummy.service",
"memory"))
CGroupConfiguratorSystemdTestCase._get_new_cgroup_configurator_instance(
).disable()
self.assertEquals(len(CGroupsTelemetry._tracked), 0)
def test_send_extension_metrics_telemetry_for_unsupported_cgroup(
self, patch_periodic_warn, patch_add_metric, *args):
CGroupsTelemetry._tracked.append(
CGroup("cgroup_name", "/test/path", "io"))
PollResourceUsageOperation().run()
self.assertEqual(1, patch_periodic_warn.call_count)
self.assertEqual(
0, patch_add_metric.call_count) # No metrics should be sent.
def test_send_extension_metrics_telemetry_for_unsupported_cgroup(
self, patch_periodic_warn, patch_add_metric, *args):
CGroupsTelemetry._tracked.append(
CGroup("cgroup_name", "/test/path", "io"))
monitor_handler = get_monitor_handler()
monitor_handler.init_protocols()
monitor_handler.last_cgroup_polling_telemetry = datetime.datetime.utcnow(
) - timedelta(hours=1)
monitor_handler.last_cgroup_report_telemetry = datetime.datetime.utcnow(
) - timedelta(hours=1)
monitor_handler.poll_telemetry_metrics()
self.assertEqual(1, patch_periodic_warn.call_count)
self.assertEqual(
0, patch_add_metric.call_count) # No metrics should be sent.
monitor_handler.stop()