Retrieve latest container id from cache when sending telemetry events (…

…Azure#1628)

azurelinuxagent/common/event.py

@@ -28,6 +28,7 @@
 import azurelinuxagent.common.logger as logger
 from azurelinuxagent.common.exception import EventError
 from azurelinuxagent.common.future import ustr
+from azurelinuxagent.common.protocol.wire import CONTAINER_ID_ENV_VARIABLE
 from azurelinuxagent.common.protocol.restapi import TelemetryEventParam, \
     TelemetryEvent, \
     get_properties
@@ -94,6 +95,7 @@ class WALAEventOperation:
     WALAEventOperation.UnInstall,
 ]
 
+
 class EventStatus(object):
     EVENT_STATUS_FILE = "event_status.json"
 
@@ -277,6 +279,8 @@ def _add_event(self, duration, evt_type, is_internal, is_success, message, name,
         event.parameters.append(TelemetryEventParam('Message', message))
         event.parameters.append(TelemetryEventParam('Duration', duration))
         event.parameters.append(TelemetryEventParam('ExtensionType', evt_type))
+        event.parameters.append(TelemetryEventParam('ContainerId',
+                                                    os.environ.get(CONTAINER_ID_ENV_VARIABLE, "UNINITIALIZED")))
 
         data = get_properties(event)
         try:

azurelinuxagent/common/protocol/wire.py

@@ -56,6 +56,8 @@
 AGENTS_MANIFEST_FILE_NAME = "{0}.{1}.agentsManifest"
 TRANSPORT_CERT_FILE_NAME = "TransportCert.pem"
 TRANSPORT_PRV_FILE_NAME = "TransportPrivate.pem"
+# Store the last retrieved container id as an environment variable to be shared between threads for telemetry purposes
+CONTAINER_ID_ENV_VARIABLE = "AZURE_GUEST_AGENT_CONTAINER_ID"
 
 PROTOCOL_VERSION = "2012-11-30"
 ENDPOINT_FINE_NAME = "WireServer"
@@ -112,7 +114,6 @@ def get_vminfo(self):
         vminfo.tenantName = hosting_env.deployment_name
         vminfo.roleName = hosting_env.role_name
         vminfo.roleInstanceName = goal_state.role_instance_id
-        vminfo.containerId = goal_state.container_id
         return vminfo
 
     def get_certs(self):
@@ -1352,6 +1353,7 @@ def parse(self, xml_text):
         self.role_config_name = findtext(role_config, "ConfigName")
         container = find(xml_doc, "Container")
         self.container_id = findtext(container, "ContainerId")
+        os.environ[CONTAINER_ID_ENV_VARIABLE] = self.container_id
         self.remote_access_uri = findtext(container, "RemoteAccessInfo")
         lbprobe_ports = find(xml_doc, "LBProbePorts")
         self.load_balancer_probe_port = findtext(lbprobe_ports, "Port")

azurelinuxagent/ga/monitor.py

@@ -188,8 +188,6 @@ def init_sysinfo(self):
                                                     vminfo.roleName))
             self.sysinfo.append(TelemetryEventParam("RoleInstanceName",
                                                     vminfo.roleInstanceName))
-            self.sysinfo.append(TelemetryEventParam("ContainerId",
-                                                    vminfo.containerId))
         except ProtocolError as e:
             logger.warn("Failed to get system info: {0}", ustr(e))
 

tests/common/test_event.py

@@ -25,6 +25,7 @@
 from azurelinuxagent.common.exception import EventError
 from azurelinuxagent.common.future import ustr
 from azurelinuxagent.common.utils.extensionprocessutil import read_output
+from azurelinuxagent.common.protocol.wire import GoalState
 from azurelinuxagent.common.version import CURRENT_VERSION
 from azurelinuxagent.ga.monitor import MonitorHandler
 
@@ -34,6 +35,55 @@
 
 
 class TestEvent(AgentTestCase):
+    def test_add_event_should_read_container_id_from_process_environment(self):
+        tmp_file = os.path.join(self.tmp_dir, "tmp_file")
+
+        def patch_save_event(json_data):
+            fileutil.write_file(tmp_file, json_data)
+
+        with patch("azurelinuxagent.common.event.EventLogger.save_event", side_effect=patch_save_event):
+            # No container id is set
+            os.environ.pop(event.CONTAINER_ID_ENV_VARIABLE, None)
+            event.add_event(name='dummy_name')
+            data = fileutil.read_file(tmp_file)
+            self.assertTrue('{"name": "ContainerId", "value": "UNINITIALIZED"}' in data or
+                            '{"value": "UNINITIALIZED", "name": "ContainerId"}' in data)
+
+            # Container id is set as an environment variable explicitly
+            os.environ[event.CONTAINER_ID_ENV_VARIABLE] = '424242'
+            event.add_event(name='dummy_name')
+            data = fileutil.read_file(tmp_file)
+            self.assertTrue('{{"name": "ContainerId", "value": "{0}"}}'.format(
+                                os.environ[event.CONTAINER_ID_ENV_VARIABLE]) in data or
+                            '{{"value": "{0}", "name": "ContainerId"}}'.format(
+                                os.environ[event.CONTAINER_ID_ENV_VARIABLE]) in data)
+
+            # Container id is set as an environment variable when parsing the goal state
+            xml_text = load_data("wire/goal_state.xml")
+            goal_state = GoalState(xml_text)
+
+            container_id = goal_state.container_id
+            event.add_event(name='dummy_name')
+            data = fileutil.read_file(tmp_file)
+            self.assertTrue('{{"name": "ContainerId", "value": "{0}"}}'.format(container_id) in data or
+                            '{{"value": "{0}", "name": "ContainerId"}}'.format(container_id), data)
+
+            # Container id is updated as the goal state changes, both in telemetry event and in environment variables
+            new_container_id = "z6d5526c-5ac2-4200-b6e2-56f2b70c5ab2"
+            xml_text = load_data("wire/goal_state.xml")
+            xml_text_updated = xml_text.replace("c6d5526c-5ac2-4200-b6e2-56f2b70c5ab2", new_container_id)
+            goal_state = GoalState(xml_text_updated)
+
+            event.add_event(name='dummy_name')
+            data = fileutil.read_file(tmp_file)
+
+            # Assert both the environment variable and telemetry event got updated
+            self.assertEquals(os.environ[event.CONTAINER_ID_ENV_VARIABLE], new_container_id)
+            self.assertTrue('{{"name": "ContainerId", "value": "{0}"}}'.format(new_container_id) in data or
+                            '{{"value": "{0}", "name": "ContainerId"}}'.format(new_container_id), data)
+
+        os.environ.pop(event.CONTAINER_ID_ENV_VARIABLE)
+
     def test_event_status_event_marked(self):
         es = event.__event_status__
 
@@ -239,7 +289,7 @@ def test_save_event_message_with_non_ascii_characters(self):
             event_str = MonitorHandler.collect_event(os.path.join(self.tmp_dir, tld_file))
             event_json = json.loads(event_str)
 
-            self.assertEqual(len(event_json["parameters"]), 8)
+            self.assertEqual(len(event_json["parameters"]), 9)
 
             for i in event_json["parameters"]:
                 if i["name"] == "Name":

tests/ga/test_extension.py

@@ -27,6 +27,7 @@
 from azurelinuxagent.ga.exthandlers import *
 from azurelinuxagent.common.protocol.wire import WireProtocol, InVMArtifactsProfile
 
+# Mock sleep to reduce test execution time
 SLEEP = time.sleep
 
 

tests/ga/test_monitor.py

@@ -97,19 +97,16 @@ def test_add_sysinfo(self, *args):
         tenant_name = 'dummy_tenant'
         role_name = 'dummy_role'
         role_instance_name = 'dummy_role_instance'
-        container_id = 'dummy_container_id'
 
         vm_name_param = "VMName"
         tenant_name_param = "TenantName"
         role_name_param = "RoleName"
         role_instance_name_param = "RoleInstanceName"
-        container_id_param = "ContainerId"
 
         sysinfo = [TelemetryEventParam(vm_name_param, vm_name),
                    TelemetryEventParam(tenant_name_param, tenant_name),
                    TelemetryEventParam(role_name_param, role_name),
-                   TelemetryEventParam(role_instance_name_param, role_instance_name),
-                   TelemetryEventParam(container_id_param, container_id)]
+                   TelemetryEventParam(role_instance_name_param, role_instance_name)]
         monitor_handler.sysinfo = sysinfo
         monitor_handler.add_sysinfo(event)
 
@@ -130,11 +127,8 @@ def test_add_sysinfo(self, *args):
             elif p.name == role_instance_name_param:
                 self.assertEqual(role_instance_name, p.value)
                 counter += 1
-            elif p.name == container_id_param:
-                self.assertEqual(container_id, p.value)
-                counter += 1
 
-        self.assertEqual(5, counter)
+        self.assertEqual(4, counter)
 
     @patch("azurelinuxagent.ga.monitor.MonitorHandler.send_telemetry_heartbeat")
     @patch("azurelinuxagent.ga.monitor.MonitorHandler.collect_and_send_events")