added retries for agent cgroups test

tests_e2e/tests/lib/cgroup_helpers.py

@@ -8,6 +8,7 @@
 from azurelinuxagent.common.version import DISTRO_NAME, DISTRO_VERSION
 from tests_e2e.tests.lib.agent_log import AgentLog
 from tests_e2e.tests.lib.logging import log
+from tests_e2e.tests.lib.retry import retry_if_false
 
 BASE_CGROUP = '/sys/fs/cgroup'
 AGENT_CGROUP_NAME = 'WALinuxAgent'
@@ -93,23 +94,26 @@ def verify_agent_cgroup_assigned_correctly():
     This method checks agent is running and assigned to the correct cgroup using service status output
     """
     log.info("===== Verifying the daemon and the agent are assigned to the same correct cgroup using systemd")
-    service_status = shellutil.run_command(["systemctl", "status", systemd.get_agent_unit_name()])
-    log.info("Agent service status output:\n%s", service_status)
-    is_active = False
-    is_cgroup_assigned = False
     cgroup_mount_path = get_agent_cgroup_mount_path()
-    is_active_pattern = re.compile(r".*Active:\s+active.*")
-
-    for line in service_status.splitlines():
-        if re.match(is_active_pattern, line):
-            is_active = True
-        elif cgroup_mount_path in line:
-            is_cgroup_assigned = True
-
-    if not is_active:
-        fail('walinuxagent service was not active/running. Service status:{0}'.format(service_status))
-    if not is_cgroup_assigned:
-        fail('walinuxagent service was not assigned to the expected cgroup:{0}'.format(cgroup_mount_path))
+    service_status = ""
+
+    def check_agent_service_cgroup():
+        is_active = False
+        is_cgroup_assigned = False
+        service_status = shellutil.run_command(["systemctl", "status", systemd.get_agent_unit_name()])
+        log.info("Agent service status output:\n%s", service_status)
+        is_active_pattern = re.compile(r".*Active:\s+active.*")
+
+        for line in service_status.splitlines():
+            if re.match(is_active_pattern, line):
+                is_active = True
+            elif cgroup_mount_path in line:
+                is_cgroup_assigned = True
+
+        return is_active and is_cgroup_assigned
+
+    if not retry_if_false(check_agent_service_cgroup):
+        fail('walinuxagent service was not assigned to the expected cgroup:{0}. Current agent status:{1}'.format(cgroup_mount_path, service_status))
 
     log.info("Successfully verified the agent cgroup assigned correctly by systemd\n")
 

tests_e2e/tests/scripts/agent_cgroups-check_cgroups_agent.py

@@ -61,22 +61,25 @@ def verify_agent_cgroup_created_on_file_system():
     """
     log.info("===== Verifying the agent cgroup paths exist on file system")
     agent_cgroup_mount_path = get_agent_cgroup_mount_path()
-    all_agent_cgroup_controllers_path_exist = True
+    log.info("expected agent cgroup mount path: %s", agent_cgroup_mount_path)
+
     missing_agent_cgroup_controllers_path = []
     verified_agent_cgroup_controllers_path = []
 
-    log.info("expected agent cgroup mount path: %s", agent_cgroup_mount_path)
+    def is_agent_cgroup_controllers_path_exist():
+        all_controllers_path_exist = True
 
-    for controller in AGENT_CONTROLLERS:
-        agent_controller_path = os.path.join(BASE_CGROUP, controller, agent_cgroup_mount_path[1:])
+        for controller in AGENT_CONTROLLERS:
+            agent_controller_path = os.path.join(BASE_CGROUP, controller, agent_cgroup_mount_path[1:])
 
-        if not os.path.exists(agent_controller_path):
-            all_agent_cgroup_controllers_path_exist = False
-            missing_agent_cgroup_controllers_path.append(agent_controller_path)
-        else:
-            verified_agent_cgroup_controllers_path.append(agent_controller_path)
+            if not os.path.exists(agent_controller_path):
+                all_controllers_path_exist = False
+                missing_agent_cgroup_controllers_path.append(agent_controller_path)
+            else:
+                verified_agent_cgroup_controllers_path.append(agent_controller_path)
+        return all_controllers_path_exist
 
-    if not all_agent_cgroup_controllers_path_exist:
+    if not retry_if_false(is_agent_cgroup_controllers_path_exist):
         fail("Agent's cgroup paths couldn't be found on file system. Missing agent cgroups path :{0}.\n Verified agent cgroups path:{1}".format(missing_agent_cgroup_controllers_path, verified_agent_cgroup_controllers_path))
 
     log.info('Verified all agent cgroup paths are present.\n {0}'.format(verified_agent_cgroup_controllers_path))