Added a 256kb custom properties truncation limit on specific GenAI attributes (#45749)

* Added a 256kb custom properties truncation limit on specific GenAI attributes

* Update CHANGELOG

* Move max_length declaration to the top

* Retrigger CI/CD pipeline

* Fix cspell error

* Retrigger CI/CD pipeline

* Retrigger CI/CD pipeline

* Retrigger CI/CD pipeline

* Retrigger CI/CD pipeline

Author: rads-1996

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

@@ -3,7 +3,8 @@
 ## 1.0.0b49 (Unreleased)
 
 ### Features Added
-
+- Added a 256kb custom properties truncation limit on specific GenAI attributes
+  ([#45749](https://github.com/Azure/azure-sdk-for-python/pull/45749))
 - Add Browser SDK loader SDK Stats feature bit ([#42904](https://github.com/Azure/azure-sdk-for-python/pull/44162))
 
 ### Breaking Changes

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

@@ -352,6 +352,7 @@ def _is_any_synthetic_source(properties: Optional[Any]) -> bool:
 # pylint: disable=W0622
 def _filter_custom_properties(properties: Attributes, filter=None) -> Dict[str, str]:
     max_length = 64 * 1024
+    max_length_for_gen_ai_attributes = 256 * 1024
     processed_properties: Dict[str, str] = {}
     if not properties:
         return processed_properties
@@ -365,7 +366,7 @@ def _filter_custom_properties(properties: Attributes, filter=None) -> Dict[str,
         if not key or len(key) > 150 or val is None:
             continue
         if key in _GEN_AI_ATTRIBUTES:
-            processed_properties[key] = str(val)
+            processed_properties[key] = str(val)[:max_length_for_gen_ai_attributes]
         else:
             processed_properties[key] = str(val)[:max_length]
     return processed_properties

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

@@ -54,8 +54,8 @@ def test_filter_custom_properties_drops_invalid_entries(self):
         self.assertEqual(filtered["short"], "ok")
         self.assertNotIn("k" * 151, filtered)
 
-    def test_custom_properties_gen_ai_attributes_not_truncated(self):
-        # All values in _GEN_AI_ATTRIBUTES should not be truncated even when > 64KiB
+    def test_custom_properties_gen_ai_attributes_not_truncated_at_64kb(self):
+        # All values in _GEN_AI_ATTRIBUTES should not be truncated at > 64kb but at > 256kb
         large_value = "x" * (64 * 1024 + 1000)
         properties = {key: large_value for key in _GEN_AI_ATTRIBUTES}
         filtered = _utils._filter_custom_properties(properties)
@@ -65,7 +65,7 @@ def test_custom_properties_gen_ai_attributes_not_truncated(self):
                 self.assertEqual(len(filtered[key]), 64 * 1024 + 1000)
 
     def test_filter_custom_properties_non_gen_ai_truncated_at_64kb(self):
-        # Regular properties exceeding 64KiB should be truncated
+        # Regular properties exceeding 64kb should be truncated
         max_length = 64 * 1024
         large_value = "y" * (max_length + 2000)
         properties = {
@@ -81,9 +81,10 @@ def test_filter_custom_properties_non_gen_ai_truncated_at_64kb(self):
                 self.assertEqual(len(filtered[key]), max_length)
 
     def test_filter_custom_properties_mixed_gen_ai_and_regular(self):
-        # Gen AI attributes keep full value, regular ones are truncated
+        # Gen AI attributes truncated at 256kb, regular ones are truncated at 64kb
         max_length = 64 * 1024
-        large_value = "z" * (max_length + 3000)
+        max_length_for_gen_ai_attributes = 256 * 1024
+        large_value = "z" * (1024 * 1024 + 3000)
         properties = {
             "gen_ai.input.messages": large_value,
             "gen_ai.output.messages": large_value,
@@ -92,14 +93,25 @@ def test_filter_custom_properties_mixed_gen_ai_and_regular(self):
             "db.statement": large_value,
         }
         filtered = _utils._filter_custom_properties(properties)
-        # Gen AI attributes — not truncated
-        self.assertEqual(len(filtered["gen_ai.input.messages"]), max_length + 3000)
-        self.assertEqual(len(filtered["gen_ai.output.messages"]), max_length + 3000)
-        # Regular attributes — truncated
+
+        self.assertEqual(len(filtered["gen_ai.input.messages"]), max_length_for_gen_ai_attributes)
+        self.assertEqual(len(filtered["gen_ai.output.messages"]), max_length_for_gen_ai_attributes)
+
         self.assertEqual(len(filtered["gen_ai.agent.version"]), max_length)
         self.assertEqual(len(filtered["span_kind"]), max_length)
         self.assertEqual(len(filtered["db.statement"]), max_length)
 
+    def test_custom_properties_gen_ai_attributes_truncated_at_256kb(self):
+        # All values in _GEN_AI_ATTRIBUTES should be truncated when > 256kb
+        max_length_for_gen_ai_attributes = 256 * 1024
+        large_value = "x" * (256 * 1024 + 1000)
+        properties = {key: large_value for key in _GEN_AI_ATTRIBUTES}
+        filtered = _utils._filter_custom_properties(properties)
+        for key in _GEN_AI_ATTRIBUTES:
+            with self.subTest(key=key):
+                self.assertIn(key, filtered)
+                self.assertEqual(len(filtered[key]), max_length_for_gen_ai_attributes)
+
     def test_nanoseconds_to_duration(self):
         ns_to_duration = _utils.ns_to_duration
         self.assertEqual(ns_to_duration(0), "0.00:00:00.000")