Cgroup rewrite: uses systemctl for expressing desired configuration i…

…nstead drop-in files (#3269)

* cgroup rewrite

* new test

* mock cleanup

* logging

* address comments

* removed script

* new comment changes

* pylint

* e2e test changes

azurelinuxagent/agent.py

@@ -34,7 +34,7 @@
 from azurelinuxagent.ga import logcollector, cgroupconfigurator
 from azurelinuxagent.ga.cgroupcontroller import AGENT_LOG_COLLECTOR
 from azurelinuxagent.ga.cpucontroller import _CpuController
-from azurelinuxagent.ga.cgroupapi import get_cgroup_api, InvalidCgroupMountpointException
+from azurelinuxagent.ga.cgroupapi import create_cgroup_api, InvalidCgroupMountpointException
 from azurelinuxagent.ga.firewall_manager import FirewallManager
 
 import azurelinuxagent.common.conf as conf
@@ -216,7 +216,8 @@ def collect_logs(self, is_full_mode):
         tracked_controllers = []
         if CollectLogsHandler.is_enabled_monitor_cgroups_check():
             try:
-                cgroup_api = get_cgroup_api()
+                cgroup_api = create_cgroup_api()
+                logger.info("Using cgroup {0} for resource enforcement and monitoring".format(cgroup_api.get_cgroup_version()))
             except InvalidCgroupMountpointException as e:
                 event.warn(WALAEventOperation.LogCollection, "The agent does not support cgroups if the default systemd mountpoint is not being used: {0}", ustr(e))
                 sys.exit(logcollector.INVALID_CGROUPS_ERRCODE)
@@ -226,6 +227,8 @@ def collect_logs(self, is_full_mode):
 
             log_collector_cgroup = cgroup_api.get_process_cgroup(process_id="self", cgroup_name=AGENT_LOG_COLLECTOR)
             tracked_controllers = log_collector_cgroup.get_controllers()
+            for controller in tracked_controllers:
+                logger.info("{0} controller for cgroup: {1}".format(controller.get_controller_type(), controller))
 
             if len(tracked_controllers) != len(log_collector_cgroup.get_supported_controller_names()):
                 event.warn(WALAEventOperation.LogCollection, "At least one required controller is missing. The following controllers are required for the log collector to run: {0}", log_collector_cgroup.get_supported_controller_names())

azurelinuxagent/common/osutil/systemd.py

@@ -43,7 +43,13 @@ def get_version():
     #    systemd 245 (245.4-4ubuntu3)
     #    +PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP etc
     #
-    return shellutil.run_command(['systemctl', '--version'])
+    # return fist line systemd 245 (245.4-4ubuntu3)
+    try:
+        output = shellutil.run_command(['systemctl', '--version'])
+        version = output.split('\n')[0]
+        return version
+    except Exception:
+        return "unknown"
 
 
 def get_unit_file_install_path():
@@ -84,3 +90,44 @@ def get_unit_property(unit_name, property_name):
         raise ValueError("Can't find property {0} of {1}".format(property_name, unit_name))
     return match.group('value')
 
+
+def set_unit_run_time_property(unit_name, property_name, value):
+    """
+    Set a property of a unit at runtime
+
+    Note: --runtime settings only apply until the next reboot
+    """
+    try:
+        # Ex: systemctl set-property foobar.service CPUWeight=200 --runtime
+        shellutil.run_command(["systemctl", "set-property", unit_name, "{0}={1}".format(property_name, value), "--runtime"])
+    except shellutil.CommandError as e:
+        raise ValueError("Can't set property {0} of {1}: {2}".format(property_name, unit_name, e))
+
+
+def set_unit_run_time_properties(unit_name, property_names, values):
+    """
+    Set multiple properties of a unit at runtime
+
+    Note: --runtime settings only apply until the next reboot
+    """
+    if len(property_names) != len(values):
+        raise ValueError("The number of property names:{0} and values:{1} must be the same".format(property_names, values))
+
+    properties = ["{0}={1}".format(name, value) for name, value in zip(property_names, values)]
+
+    try:
+        # Ex: systemctl set-property foobar.service CPUWeight=200 MemoryMax=2G IPAccounting=yes --runtime
+        shellutil.run_command(["systemctl", "set-property", unit_name] + properties + ["--runtime"])
+    except shellutil.CommandError as e:
+        raise ValueError("Can't set properties {0} of {1}: {2}".format(properties, unit_name, e))
+
+
+def is_unit_loaded(unit_name):
+    """
+    Determine if a unit is loaded
+    """
+    try:
+        value = get_unit_property(unit_name, "LoadState")
+        return value.lower() == "loaded"
+    except shellutil.CommandError:
+        return False

azurelinuxagent/ga/cgroupapi.py

@@ -59,7 +59,7 @@ class CGroupUtil(object):
     Cgroup utility methods which are independent of systemd cgroup api.
     """
     @staticmethod
-    def cgroups_supported():
+    def distro_supported():
         distro_info = get_distro()
         distro_name = distro_info[0]
         try:
@@ -149,7 +149,7 @@ def __init__(self, msg=None):
         super(InvalidCgroupMountpointException, self).__init__(msg)
 
 
-def get_cgroup_api():
+def create_cgroup_api():
     """
     Determines which version of Cgroup should be used for resource enforcement and monitoring by the Agent and returns
     the corresponding Api.
@@ -172,7 +172,6 @@ def get_cgroup_api():
     root_hierarchy_mode = shellutil.run_command(["stat", "-f", "--format=%T", CGROUP_FILE_SYSTEM_ROOT]).rstrip()
 
     if root_hierarchy_mode == "cgroup2fs":
-        log_cgroup_info("Using cgroup v2 for resource enforcement and monitoring")
         return SystemdCgroupApiv2()
 
     elif root_hierarchy_mode == "tmpfs":
@@ -192,7 +191,6 @@ def get_cgroup_api():
         # mounted in a location other than the systemd default, raise Exception.
         if not cgroup_api_v1.are_mountpoints_systemd_created():
             raise InvalidCgroupMountpointException("Expected cgroup controllers to be mounted at '{0}', but at least one is not. v1 mount points: \n{1}".format(CGROUP_FILE_SYSTEM_ROOT, json.dumps(cgroup_api_v1.get_controller_mountpoints())))
-        log_cgroup_info("Using cgroup v1 for resource enforcement and monitoring")
         return cgroup_api_v1
 
     raise CGroupsException("{0} has an unexpected file type: {1}".format(CGROUP_FILE_SYSTEM_ROOT, root_hierarchy_mode))
@@ -206,6 +204,12 @@ def __init__(self):
         self._systemd_run_commands = []
         self._systemd_run_commands_lock = threading.RLock()
 
+    def get_cgroup_version(self):
+        """
+        Returns the version of the cgroup hierarchy in use.
+        """
+        return NotImplementedError()
+
     def get_systemd_run_commands(self):
         """
         Returns a list of the systemd-run commands currently running (given as PIDs)
@@ -297,6 +301,12 @@ def _get_controller_mountpoints():
                     mount_points[controller] = path
         return mount_points
 
+    def get_cgroup_version(self):
+        """
+        Returns the version of the cgroup hierarchy in use.
+        """
+        return "v1"
+
     def get_controller_mountpoints(self):
         """
         Returns a dictionary of controller-mountpoint mappings.
@@ -479,6 +489,12 @@ def _get_root_cgroup_path():
                     return root_cgroup_path
         return ""
 
+    def get_cgroup_version(self):
+        """
+        Returns the version of the cgroup hierarchy in use.
+        """
+        return "v2"
+
     def get_root_cgroup_path(self):
         """
         Returns the unified cgroup mountpoint.
@@ -650,8 +666,6 @@ def get_controllers(self, expected_relative_path=None):
                 controller = MemoryControllerV1(self._cgroup_name, controller_path)
 
             if controller is not None:
-                msg = "{0} controller for cgroup: {1}".format(supported_controller_name, controller)
-                log_cgroup_info(msg)
                 controllers.append(controller)
 
         return controllers
@@ -729,8 +743,6 @@ def get_controllers(self, expected_relative_path=None):
                 controller = MemoryControllerV2(self._cgroup_name, self._cgroup_path)
 
             if controller is not None:
-                msg = "{0} controller for cgroup: {1}".format(supported_controller_name, controller)
-                log_cgroup_info(msg)
                 controllers.append(controller)
 
         return controllers