Update AKS checks and tests to use KUBERNETES_SERVICE_HOST

sdk/monitor/azure-monitor-opentelemetry-exporter/CHANGELOG.md

@@ -23,6 +23,8 @@
   ([#39886](https://github.com/Azure/azure-sdk-for-python/pull/39886))
 - Populate `client_Ip` on `customEvent` telemetry
   ([#39923](https://github.com/Azure/azure-sdk-for-python/pull/39923))
+- Update AKS check to use KUBERNETES_SERVICE_HOST
+  ([#39941](https://github.com/Azure/azure-sdk-for-python/pull/39941))
 
 ### Bugs Fixed
 

sdk/monitor/azure-monitor-opentelemetry-exporter/azure/monitor/opentelemetry/exporter/_constants.py

@@ -24,6 +24,7 @@
 _FUNCTIONS_WORKER_RUNTIME = "FUNCTIONS_WORKER_RUNTIME"
 _PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY = "PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY"
 _AKS_ARM_NAMESPACE_ID = "AKS_ARM_NAMESPACE_ID"
+_KUBERNETES_SERVICE_HOST = "KUBERNETES_SERVICE_HOST"
 
 # Network
 

sdk/monitor/azure-monitor-opentelemetry-exporter/azure/monitor/opentelemetry/exporter/_utils.py

@@ -23,8 +23,9 @@
 from azure.monitor.opentelemetry.exporter._constants import (
     _AKS_ARM_NAMESPACE_ID,
     _DEFAULT_AAD_SCOPE,
-    _INSTRUMENTATIONS_BIT_MAP,
     _FUNCTIONS_WORKER_RUNTIME,
+    _INSTRUMENTATIONS_BIT_MAP,
+    _KUBERNETES_SERVICE_HOST,
     _PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY,
     _WEBSITE_SITE_NAME,
 )
@@ -63,7 +64,7 @@ def _is_on_functions():
 
 
 def _is_on_aks():
-    return _AKS_ARM_NAMESPACE_ID in environ
+    return _AKS_ARM_NAMESPACE_ID in environ or _KUBERNETES_SERVICE_HOST in environ
 
 
 # Attach
@@ -74,6 +75,8 @@ def _is_attach_enabled():
         return isdir("/agents/python/")
     if _is_on_functions():
         return environ.get(_PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY) == "true"
+    if _is_on_aks():
+        return _AKS_ARM_NAMESPACE_ID in environ
     return False
 
 

sdk/monitor/azure-monitor-opentelemetry-exporter/azure/monitor/opentelemetry/exporter/statsbeat/_statsbeat_metrics.py

@@ -18,6 +18,7 @@
 from azure.monitor.opentelemetry.exporter._constants import (
     _ATTACH_METRIC_NAME,
     _FEATURE_METRIC_NAME,
+    _KUBERNETES_SERVICE_HOST,
     _REQ_DURATION_NAME,
     _REQ_EXCEPTION_NAME,
     _REQ_FAILURE_NAME,
@@ -187,7 +188,10 @@ def _get_attach_metric(self, options: CallbackOptions) -> Iterable[Observation]:
         elif _utils._is_on_aks():
             # AKS
             rp = _RP_Names.AKS.value
-            rpId = os.environ.get(_AKS_ARM_NAMESPACE_ID, "")
+            if _AKS_ARM_NAMESPACE_ID in os.environ:
+                rpId = os.environ.get(_AKS_ARM_NAMESPACE_ID, "")
+            else:
+                rpId = os.environ.get(_KUBERNETES_SERVICE_HOST , "")
         elif self._vm_retry and self._get_azure_compute_metadata():
             # VM
             rp = _RP_Names.VM.value

sdk/monitor/azure-monitor-opentelemetry-exporter/tests/statsbeat/test_statsbeat.py

@@ -447,9 +447,10 @@ def test_get_attach_metric_functions(self, metadata_mock):
         os.environ,
         {
             "AKS_ARM_NAMESPACE_ID": "namespace_id",
+            "KUBERNETES_SERVICE_HOST": "TEST_KUBERNETES_SERVICE_HOST",
         },
     )
-    def test_get_attach_metric_aks(self):
+    def test_get_attach_metric_aks_attach(self):
         attributes = dict(_StatsbeatMetrics._COMMON_ATTRIBUTES)
         self.assertEqual(attributes["rp"], _RP_Names.UNKNOWN.value)
         attributes["rp"] = _RP_Names.AKS.value
@@ -460,6 +461,24 @@ def test_get_attach_metric_aks(self):
             self.assertEqual(obs.attributes, attributes)
         self.assertEqual(_StatsbeatMetrics._COMMON_ATTRIBUTES["rp"], _RP_Names.AKS.value)
 
+    # pylint: disable=protected-access
+    @mock.patch.dict(
+        os.environ,
+        {
+            "KUBERNETES_SERVICE_HOST": "TEST_KUBERNETES_SERVICE_HOST",
+        },
+    )
+    def test_get_attach_metric_aks_manual(self):
+        attributes = dict(_StatsbeatMetrics._COMMON_ATTRIBUTES)
+        self.assertEqual(attributes["rp"], _RP_Names.UNKNOWN.value)
+        attributes["rp"] = _RP_Names.AKS.value
+        attributes["rpId"] = "TEST_KUBERNETES_SERVICE_HOST"
+        observations = self._metric._get_attach_metric(options=None)
+        for obs in observations:
+            self.assertEqual(obs.value, 1)
+            self.assertEqual(obs.attributes, attributes)
+        self.assertEqual(_StatsbeatMetrics._COMMON_ATTRIBUTES["rp"], _RP_Names.AKS.value)
+
     @mock.patch(
         "azure.monitor.opentelemetry.exporter.statsbeat._statsbeat_metrics._StatsbeatMetrics._get_azure_compute_metadata"
     )

sdk/monitor/azure-monitor-opentelemetry-exporter/tests/test_utils.py

@@ -29,6 +29,8 @@
 TEST_TIMESTAMP = "TEST_TIMESTAMP"
 TEST_TIME = "TEST_TIME"
 TEST_WEBSITE_SITE_NAME = "TEST_WEBSITE_SITE_NAME"
+TEST_KUBERNETES_SERVICE_HOST = "TEST_KUBERNETES_SERVICE_HOST"
+TEST_AKS_ARM_NAMESPACE_ID = "TEST_AKS_ARM_NAMESPACE_ID"
 
 
 class TestUtils(unittest.TestCase):
@@ -105,37 +107,37 @@ def test_create_telemetry_item(self, mock_ns_to_iso_str):
 
     @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
-    def test_get_sdk_version_prefix(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "uum_")
 
     @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
-    def test_get_sdk_version_prefix_linux(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_linux(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "ulm_")
 
     @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
-    def test_get_sdk_version_prefix_windows(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_windows(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "uwm_")
 
     @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
-    def test_get_sdk_version_prefix_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_attach(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "uui_")
 
     @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
-    def test_get_sdk_version_prefix_attach_linux(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_attach_linux(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "uli_")
 
     @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
-    def test_get_sdk_version_prefix_attach_windows(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_attach_windows(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "uwi_")
 
@@ -146,7 +148,7 @@ def test_get_sdk_version_prefix_attach_windows(self, mock_system, mock_getenv):
     )
     @patch("azure.monitor.opentelemetry.exporter._utils.isdir", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
-    def test_get_sdk_version_prefix_app_service(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_app_service(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "aum_")
 
@@ -155,7 +157,7 @@ def test_get_sdk_version_prefix_app_service(self, mock_system, mock_getenv):
     )
     @patch("azure.monitor.opentelemetry.exporter._utils.isdir", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
-    def test_get_sdk_version_prefix_app_service_linux(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_app_service_linux(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "alm_")
 
@@ -164,7 +166,7 @@ def test_get_sdk_version_prefix_app_service_linux(self, mock_system, mock_getenv
     )
     @patch("azure.monitor.opentelemetry.exporter._utils.isdir", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
-    def test_get_sdk_version_prefix_app_service_windows(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_app_service_windows(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "awm_")
 
@@ -173,7 +175,7 @@ def test_get_sdk_version_prefix_app_service_windows(self, mock_system, mock_gete
     )
     @patch("azure.monitor.opentelemetry.exporter._utils.isdir", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
-    def test_get_sdk_version_prefix_app_service_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_app_service_attach(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "aui_")
 
@@ -182,7 +184,7 @@ def test_get_sdk_version_prefix_app_service_attach(self, mock_system, mock_geten
     )
     @patch("azure.monitor.opentelemetry.exporter._utils.isdir", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
-    def test_get_sdk_version_prefix_app_service_linux_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_app_service_linux_attach(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "ali_")
 
@@ -191,7 +193,7 @@ def test_get_sdk_version_prefix_app_service_linux_attach(self, mock_system, mock
     )
     @patch("azure.monitor.opentelemetry.exporter._utils.isdir", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
-    def test_get_sdk_version_prefix_app_service_windows_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_app_service_windows_attach(self, mock_system, mock_isdir):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "awi_")
 
@@ -202,9 +204,8 @@ def test_get_sdk_version_prefix_app_service_windows_attach(self, mock_system, mo
         {"FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME},
         clear=True,
     )
-    @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
-    def test_get_sdk_version_prefix_function(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_function(self, mock_system):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "fum_")
 
@@ -213,9 +214,8 @@ def test_get_sdk_version_prefix_function(self, mock_system, mock_getenv):
         {"FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME},
         clear=True,
     )
-    @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
-    def test_get_sdk_version_prefix_function_linux(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_function_linux(self, mock_system):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "flm_")
 
@@ -224,45 +224,124 @@ def test_get_sdk_version_prefix_function_linux(self, mock_system, mock_getenv):
         {"FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME},
         clear=True,
     )
-    @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=False)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
-    def test_get_sdk_version_prefix_function_windows(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_function_windows(self, mock_system):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "fwm_")
 
     @patch.dict(
         "azure.monitor.opentelemetry.exporter._utils.environ",
-        {"FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME},
+        {
+            "FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME,
+            "PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY": "true",
+        },
         clear=True,
     )
-    @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
-    def test_get_sdk_version_prefix_function_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_function_attach(self, mock_system):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "fui_")
 
     @patch.dict(
         "azure.monitor.opentelemetry.exporter._utils.environ",
-        {"FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME},
+        {
+            "FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME,
+            "PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY": "true",
+        },
         clear=True,
     )
-    @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
-    def test_get_sdk_version_prefix_function_linux_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_function_linux_attach(self, mock_system):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "fli_")
 
     @patch.dict(
         "azure.monitor.opentelemetry.exporter._utils.environ",
-        {"FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME},
+        {
+            "FUNCTIONS_WORKER_RUNTIME": TEST_WEBSITE_SITE_NAME,
+            "PYTHON_APPLICATIONINSIGHTS_ENABLE_TELEMETRY": "true",
+        },
         clear=True,
     )
-    @patch("azure.monitor.opentelemetry.exporter._utils._is_attach_enabled", return_value=True)
     @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
-    def test_get_sdk_version_prefix_function_windows_attach(self, mock_system, mock_getenv):
+    def test_get_sdk_version_prefix_function_windows_attach(self, mock_system):
         result = _utils._get_sdk_version_prefix()
         self.assertEqual(result, "fwi_")
 
+    # AKS SDK Version Prefix
+
+    @patch.dict(
+        "azure.monitor.opentelemetry.exporter._utils.environ",
+        {
+            "KUBERNETES_SERVICE_HOST": TEST_KUBERNETES_SERVICE_HOST,
+        },
+        clear=True,
+    )
+    @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
+    def test_get_sdk_version_prefix_aks(self, mock_system):
+        result = _utils._get_sdk_version_prefix()
+        self.assertEqual(result, "kum_")
+
+    @patch.dict(
+        "azure.monitor.opentelemetry.exporter._utils.environ",
+        {
+            "KUBERNETES_SERVICE_HOST": TEST_KUBERNETES_SERVICE_HOST,
+        },
+        clear=True,
+    )
+    @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
+    def test_get_sdk_version_prefix_aks_linux(self, mock_system):
+        result = _utils._get_sdk_version_prefix()
+        self.assertEqual(result, "klm_")
+
+    @patch.dict(
+        "azure.monitor.opentelemetry.exporter._utils.environ",
+        {
+            "KUBERNETES_SERVICE_HOST": TEST_KUBERNETES_SERVICE_HOST,
+        },
+        clear=True,
+    )
+    @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
+    def test_get_sdk_version_prefix_aks_windows(self, mock_system):
+        result = _utils._get_sdk_version_prefix()
+        self.assertEqual(result, "kwm_")
+
+    @patch.dict(
+        "azure.monitor.opentelemetry.exporter._utils.environ",
+        {
+            "AKS_ARM_NAMESPACE_ID": TEST_AKS_ARM_NAMESPACE_ID,
+        },
+        clear=True,
+    )
+    @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="")
+    def test_get_sdk_version_prefix_aks_attach(self, mock_system):
+        result = _utils._get_sdk_version_prefix()
+        self.assertEqual(result, "kui_")
+
+    @patch.dict(
+        "azure.monitor.opentelemetry.exporter._utils.environ",
+        {
+            "AKS_ARM_NAMESPACE_ID": TEST_AKS_ARM_NAMESPACE_ID,
+        },
+        clear=True,
+    )
+    @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Linux")
+    def test_get_sdk_version_prefix_aks_linux_attach(self, mock_system):
+        result = _utils._get_sdk_version_prefix()
+        self.assertEqual(result, "kli_")
+
+    @patch.dict(
+        "azure.monitor.opentelemetry.exporter._utils.environ",
+        {
+            "AKS_ARM_NAMESPACE_ID": TEST_AKS_ARM_NAMESPACE_ID,
+        },
+        clear=True,
+    )
+    @patch("azure.monitor.opentelemetry.exporter._utils.platform.system", return_value="Windows")
+    def test_get_sdk_version_prefix_aks_windows_attach(self, mock_system):
+        result = _utils._get_sdk_version_prefix()
+        self.assertEqual(result, "kwi_")
+
     # Attach
 
     @patch(
@@ -288,7 +367,22 @@ def test_attach_app_service_disabled(self, mock_isdir):
     @patch.dict("azure.monitor.opentelemetry.exporter._utils.environ", {}, clear=True)
     def test_attach_off_app_service_with_agent(self, mock_isdir):
         # This is not an expected scenario and just tests the default
-        self.assertEqual(_utils._is_attach_enabled(), False)
+        self.assertFalse(_utils._is_attach_enabled())
+
+    @patch.dict("azure.monitor.opentelemetry.exporter._utils.environ", {
+        "KUBERNETES_SERVICE_HOST": TEST_KUBERNETES_SERVICE_HOST,
+        "AKS_ARM_NAMESPACE_ID": TEST_AKS_ARM_NAMESPACE_ID,
+    }, clear=True)
+    def test_attach_aks(self):
+        # This is not an expected scenario and just tests the default
+        self.assertTrue(_utils._is_attach_enabled())
+
+    @patch.dict("azure.monitor.opentelemetry.exporter._utils.environ", {
+        "KUBERNETES_SERVICE_HOST": TEST_KUBERNETES_SERVICE_HOST,
+    }, clear=True)
+    def test_aks_no_attach(self):
+        # This is not an expected scenario and just tests the default
+        self.assertFalse(_utils._is_attach_enabled())
 
     # Synthetic