Address feedback

Author: rads-1996

instrumentation/opentelemetry-instrumentation-genai-langchain/src/opentelemetry/instrumentation/genai/langchain/callback_handler.py

@@ -323,17 +323,19 @@ def on_llm_end(
 
                 # Token usage (extracted regardless of whether the message
                 # produced output parts — some tool-call-only responses can
-                # still report token counts).
+                # still report token counts). Only set the counts when the
+                # provider actually reported them; defaulting to 0 would
+                # fabricate telemetry when the keys are absent.
                 if message.usage_metadata:
-                    input_tokens = message.usage_metadata.get(
-                        "input_tokens", 0
-                    )
-                    llm_invocation.input_tokens = input_tokens
+                    input_tokens = message.usage_metadata.get("input_tokens")
+                    if input_tokens is not None:
+                        llm_invocation.input_tokens = input_tokens
 
                     output_tokens = message.usage_metadata.get(
-                        "output_tokens", 0
+                        "output_tokens"
                     )
-                    llm_invocation.output_tokens = output_tokens
+                    if output_tokens is not None:
+                        llm_invocation.output_tokens = output_tokens
 
         llm_invocation.output_messages = output_messages
 

instrumentation/opentelemetry-instrumentation-genai-langchain/src/opentelemetry/instrumentation/genai/langchain/utils.py

@@ -51,7 +51,7 @@ def normalize_provider(metadata: Optional[dict[str, Any]]) -> Optional[str]:
     if not metadata:
         return None
     raw = metadata.get("ls_provider")
-    if not raw:
+    if not isinstance(raw, str) or not raw:
         return None
     return _PROVIDER_NAME_OVERRIDES.get(raw, raw)
 
@@ -216,7 +216,9 @@ def make_input_message(data: Any) -> list[InputMessage]:
     data_dict = cast(dict[str, Any], data)
     messages: Any = data_dict.get("messages")
     if messages is not None:
-        if not isinstance(messages, list):
+        if isinstance(messages, (str, bytes)) or not isinstance(
+            messages, Iterable
+        ):
             return []
         return to_input_messages(messages)
     # Fallback: serialize non-message state fields as input.
@@ -247,7 +249,11 @@ def make_output_message(data: Any) -> list[OutputMessage]:
         return []
     data_dict = cast(dict[str, Any], data)
     messages: Any = data_dict.get("messages")
-    if not isinstance(messages, list):
+    if (
+        messages is None
+        or isinstance(messages, (str, bytes))
+        or not isinstance(messages, Iterable)
+    ):
         return []
     return to_output_messages(messages)
 

instrumentation/opentelemetry-instrumentation-genai-langchain/tests/test_llm_call.py

@@ -380,7 +380,7 @@ def assert_gemini_completion_attributes(
     assert span.attributes["gen_ai.provider.name"] == "gcp.gen_ai"
 
     input_tokens = response.usage_metadata.get("input_tokens")
-    if input_tokens:
+    if input_tokens is not None:
         assert (
             input_tokens
             == span.attributes[gen_ai_attributes.GEN_AI_USAGE_INPUT_TOKENS]
@@ -391,7 +391,7 @@ def assert_gemini_completion_attributes(
         )
 
     output_tokens = response.usage_metadata.get("output_tokens")
-    if output_tokens:
+    if output_tokens is not None:
         assert (
             output_tokens
             == span.attributes[gen_ai_attributes.GEN_AI_USAGE_OUTPUT_TOKENS]