def __collect_azure_unit_telemetry():
azure_units = []
try:
units = shellutil.run_command(['systemctl', 'list-units', 'azure*', '-all'])
for line in units.split('\n'):
match = re.match(r'\s?(azure[^\s]*)\s?', line, re.IGNORECASE)
if match is not None:
azure_units.append((match.group(1), line))
except shellutil.CommandError as command_error:
_log_cgroup_warning("Failed to list systemd units: {0}", ustr(command_error))
for unit_name, unit_description in azure_units:
unit_slice = "Unknown"
try:
unit_slice = systemd.get_unit_property(unit_name, "Slice")
except Exception as exception:
_log_cgroup_warning("Failed to query Slice for {0}: {1}", unit_name, ustr(exception))
_log_cgroup_info("Found an Azure unit under slice {0}: {1}", unit_slice, unit_description)
if len(azure_units) == 0:
try:
cgroups = shellutil.run_command('systemd-cgls')
for line in cgroups.split('\n'):
if re.match(r'[^\x00-\xff]+azure\.slice\s*', line, re.UNICODE):
logger.info(ustr("Found a cgroup for azure.slice\n{0}").format(cgroups))
# Don't add the output of systemd-cgls to the telemetry, since currently it does not support Unicode
add_event(op=WALAEventOperation.CGroupsInfo, message="Found a cgroup for azure.slice")
except shellutil.CommandError as command_error:
_log_cgroup_warning("Failed to list systemd units: {0}", ustr(command_error))
def __get_agent_cgroups(self, agent_slice, cpu_controller_root, memory_controller_root):
agent_unit_name = systemd.get_agent_unit_name()
expected_relative_path = os.path.join(agent_slice, agent_unit_name)
cpu_cgroup_relative_path, memory_cgroup_relative_path = self._cgroups_api.get_process_cgroup_relative_paths("self")
if cpu_cgroup_relative_path is None:
_log_cgroup_warning("The agent's process is not within a CPU cgroup")
else:
if cpu_cgroup_relative_path == expected_relative_path:
_log_cgroup_info('CPUAccounting: {0}', systemd.get_unit_property(agent_unit_name, "CPUAccounting"))
_log_cgroup_info('CPUQuota: {0}', systemd.get_unit_property(agent_unit_name, "CPUQuotaPerSecUSec"))
else:
cpu_cgroup_relative_path = None # Set the path to None to prevent monitoring
_log_cgroup_warning(
"The Agent is not in the expected CPU cgroup; will not enable monitoring. Cgroup:[{0}] Expected:[{1}]",
cpu_cgroup_relative_path,
expected_relative_path)
if memory_cgroup_relative_path is None:
_log_cgroup_warning("The agent's process is not within a memory cgroup")
else:
if memory_cgroup_relative_path == expected_relative_path:
memory_accounting = systemd.get_unit_property(agent_unit_name, "MemoryAccounting")
_log_cgroup_info('MemoryAccounting: {0}', memory_accounting)
else:
memory_cgroup_relative_path = None # Set the path to None to prevent monitoring
_log_cgroup_info(
"The Agent is not in the expected memory cgroup; will not enable monitoring. CGroup:[{0}] Expected:[{1}]",
memory_cgroup_relative_path,
expected_relative_path)
if cpu_controller_root is not None and cpu_cgroup_relative_path is not None:
agent_cpu_cgroup_path = os.path.join(cpu_controller_root, cpu_cgroup_relative_path)
else:
agent_cpu_cgroup_path = None
if memory_controller_root is not None and memory_cgroup_relative_path is not None:
agent_memory_cgroup_path = os.path.join(memory_controller_root, memory_cgroup_relative_path)
else:
agent_memory_cgroup_path = None
return agent_cpu_cgroup_path, agent_memory_cgroup_path
def test_get_unit_property_should_return_the_value_of_the_given_property(
self):
with mock_cgroup_environment(self.tmp_dir):
cpu_accounting = systemd.get_unit_property("walinuxagent.service",
"CPUAccounting")
self.assertEqual(
cpu_accounting, "no",
"Property {0} of {1} is incorrect".format(
"CPUAccounting", "walinuxagent.service"))
def __collect_agent_unit_files_telemetry():
agent_unit_files = []
agent_service_name = get_osutil().get_service_name()
try:
fragment_path = systemd.get_unit_property(agent_service_name, "FragmentPath")
if fragment_path != "/lib/systemd/system/{0}.service".format(agent_service_name):
agent_unit_files.append(fragment_path)
except Exception as exception:
_log_cgroup_warning("Failed to query the agent's FragmentPath: {0}", ustr(exception))
try:
drop_in_paths = systemd.get_unit_property(agent_service_name, "DropInPaths")
for path in drop_in_paths.split():
agent_unit_files.append(path)
except Exception as exception:
_log_cgroup_warning("Failed to query the agent's DropInPaths: {0}", ustr(exception))
for unit_file in agent_unit_files:
try:
with open(unit_file, "r") as file_object:
_log_cgroup_info("Found a custom unit file for the agent: {0}\n{1}", unit_file, file_object.read())
except Exception as exception:
_log_cgroup_warning("Can't read {0}: {1}", unit_file, ustr(exception))
def initialize(self):
try:
if self._initialized:
return
# check whether cgroup monitoring is supported on the current distro
self._cgroups_supported = CGroupsApi.cgroups_supported()
if not self._cgroups_supported:
logger.info("Cgroup monitoring is not supported on {0}", get_distro())
return
# check that systemd is detected correctly
self._cgroups_api = SystemdCgroupsApi()
if not systemd.is_systemd():
_log_cgroup_warning("systemd was not detected on {0}", get_distro())
return
_log_cgroup_info("systemd version: {0}", systemd.get_version())
# This is temporarily disabled while we analyze telemetry. Likely it will be removed.
# self.__collect_azure_unit_telemetry()
# self.__collect_agent_unit_files_telemetry()
if not self.__check_no_legacy_cgroups():
return
agent_unit_name = systemd.get_agent_unit_name()
agent_slice = systemd.get_unit_property(agent_unit_name, "Slice")
if agent_slice not in (_AZURE_SLICE, "system.slice"):
_log_cgroup_warning("The agent is within an unexpected slice: {0}", agent_slice)
return
self.__setup_azure_slice()
cpu_controller_root, memory_controller_root = self.__get_cgroup_controllers()
self._agent_cpu_cgroup_path, self._agent_memory_cgroup_path = self.__get_agent_cgroups(agent_slice, cpu_controller_root, memory_controller_root)
if self._agent_cpu_cgroup_path is not None:
_log_cgroup_info("Agent CPU cgroup: {0}", self._agent_cpu_cgroup_path)
self.enable()
CGroupsTelemetry.track_cgroup(CpuCgroup(AGENT_NAME_TELEMETRY, self._agent_cpu_cgroup_path))
_log_cgroup_info('Cgroups enabled: {0}', self._cgroups_enabled)
except Exception as exception:
_log_cgroup_warning("Error initializing cgroups: {0}", ustr(exception))
finally:
self._initialized = True
def get_unit_cgroup_paths(self, unit_name):
"""
Returns a tuple with the path of the cpu and memory cgroups for the given unit.
The values returned can be None if the controller is not mounted.
Ex: ControlGroup=/azure.slice/walinuxagent.service
controlgroup_path[1:] = azure.slice/walinuxagent.service
"""
controlgroup_path = systemd.get_unit_property(unit_name,
"ControlGroup")
cpu_mount_point, memory_mount_point = self.get_cgroup_mount_points()
cpu_cgroup_path = os.path.join(cpu_mount_point, controlgroup_path[1:]) \
if cpu_mount_point is not None else None
memory_cgroup_path = os.path.join(memory_mount_point, controlgroup_path[1:]) \
if memory_mount_point is not None else None
return cpu_cgroup_path, memory_cgroup_path