feat: add server_address on llm call traces and metrics

Author: feng-95

veadk/tracing/telemetry/attributes/extractors/llm_attributes_extractors.py

@@ -14,6 +14,7 @@
 
 import json
 
+from google.adk.agents.run_config import StreamingMode
 from veadk.tracing.telemetry.attributes.extractors.types import (
     ExtractorResponse,
     LLMAttributesParams,
@@ -507,13 +508,15 @@ def llm_gen_ai_is_streaming(params: LLMAttributesParams) -> ExtractorResponse:
     for performance analysis and debugging purposes.
 
     Args:
-        params: LLM execution parameters (currently not implemented)
+        params: LLM execution parameters
 
     Returns:
-        ExtractorResponse: Response containing None (not implemented)
+        ExtractorResponse: Response containing None
     """
-    # return params.llm_request.stream
-    return ExtractorResponse(content=None)
+    is_streaming = bool(
+        params.invocation_context.run_config and params.invocation_context.run_config.streaming_mode != StreamingMode.NONE)
+
+    return ExtractorResponse(content=is_streaming)
 
 
 def llm_gen_ai_operation_name(params: LLMAttributesParams) -> ExtractorResponse:
@@ -804,6 +807,21 @@ def llm_gen_ai_request_functions(params: LLMAttributesParams) -> ExtractorRespon
 
     return ExtractorResponse(content=functions)
 
+def llm_server_address(params: LLMAttributesParams) -> ExtractorResponse:
+    """Extract the LLM server address (model API base URL).
+
+    Returns the model API base URL configured on the current Agent.
+    If the Agent or base URL is unavailable, returns 'unknown' to
+    keep the span attribute consistent.
+
+    Args:
+        params: LLM execution parameters containing invocation context
+
+    Returns:
+        ExtractorResponse: Response containing the server address or 'unknown'
+    """
+    return ExtractorResponse(content=getattr(params.invocation_context.agent, "model_api_base", None) or "unknown")
+
 
 LLM_ATTRIBUTES = {
     # -> 1. attributes
@@ -813,6 +831,7 @@ def llm_gen_ai_request_functions(params: LLMAttributesParams) -> ExtractorRespon
     "gen_ai.request.max_tokens": llm_gen_ai_request_max_tokens,
     "gen_ai.request.temperature": llm_gen_ai_request_temperature,
     "gen_ai.request.top_p": llm_gen_ai_request_top_p,
+    "server.address": llm_server_address,
     # CozeLoop required
     "gen_ai.request.functions": llm_gen_ai_request_functions,
     # -> 1.2. response

veadk/tracing/telemetry/exporters/apmplus_exporter.py

@@ -12,11 +12,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
 import time
 from dataclasses import dataclass
 from typing import Any
 
 from google.adk.agents.invocation_context import InvocationContext
+from google.adk.agents.run_config import StreamingMode
 from google.adk.events import Event
 from google.adk.models.llm_request import LlmRequest
 from google.adk.models.llm_response import LlmResponse
@@ -34,6 +36,7 @@
 from typing_extensions import override
 
 from veadk.config import settings
+from veadk.consts import DEFAULT_MODEL_AGENT_API_BASE
 from veadk.tracing.telemetry.exporters.base_exporter import BaseExporter
 from veadk.utils.logger import get_logger
 
@@ -296,13 +299,16 @@ def record_call_llm(
             llm_request: Request object with model and parameter details
             llm_response: Response object with content and usage metadata
         """
+        is_streaming = bool(
+            invocation_context.run_config and invocation_context.run_config.streaming_mode != StreamingMode.NONE)
+        server_address = getattr(invocation_context.agent, "model_api_base", None) or "unknown"
         attributes = {
             "gen_ai_system": "volcengine",
             "gen_ai_response_model": llm_request.model,
             "gen_ai_operation_name": "chat",
             "gen_ai_operation_type": "llm",
-            "stream": "false",
-            "server_address": "api.volcengine.com",
+            "stream": is_streaming,
+            "server_address":  server_address,
         }  # required by Volcengine APMPlus
 
         if llm_response.usage_metadata:
@@ -337,7 +343,7 @@ def record_call_llm(
             if llm_response.error_code and self.chat_exception_counter:
                 exception_attributes = {
                     **attributes,
-                    "error_type": llm_response.error_message,
+                    "error_type": llm_response.error_code,
                 }
                 self.chat_exception_counter.add(1, exception_attributes)