Skip to content

Same agent Smaller bill. The tools and infrastructure to get more work from your AI. Fewer tokens. Less compute. Lower cost.

[Feat]: Refactor VertexAI OpenTelemetry Monitoring Instrumentation (#784)

Patcher9560aaa5b759
diff --git a/sdk/python/pyproject.toml b/sdk/python/pyproject.toml
index 4dde472b..b1fa22db 100644
--- a/sdk/python/pyproject.toml
+++ b/sdk/python/pyproject.toml
@@ -1,6 +1,6 @@
[tool.poetry]
name = "openlit"
-version = "1.34.20"
+version = "1.34.21"
description = "OpenTelemetry-native Auto instrumentation library for monitoring LLM Applications and GPUs, facilitating the integration of observability into your GenAI-driven projects"
authors = ["OpenLIT"]
license = "Apache-2.0"
diff --git a/sdk/python/src/openlit/instrumentation/vertexai/__init__.py b/sdk/python/src/openlit/instrumentation/vertexai/__init__.py
index a7ba0b94..5e944282 100644
--- a/sdk/python/src/openlit/instrumentation/vertexai/__init__.py
+++ b/sdk/python/src/openlit/instrumentation/vertexai/__init__.py
@@ -1,4 +1,3 @@
-# pylint: disable=useless-return, bad-staticmethod-argument, disable=duplicate-code
"""Initializer of Auto Instrumentation of VertexAI Functions"""
from typing import Collection
@@ -6,92 +5,88 @@ import importlib.metadata
from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
from wrapt import wrap_function_wrapper
-from openlit.instrumentation.vertexai.vertexai import (
- send_message
-)
-from openlit.instrumentation.vertexai.async_vertexai import (
- async_send_message
-)
-
+from openlit.instrumentation.vertexai.vertexai import send_message
+from openlit.instrumentation.vertexai.async_vertexai import async_send_message
_instruments = ("google-cloud-aiplatform >= 1.38.1",)
class VertexAIInstrumentor(BaseInstrumentor):
"""
- An instrumentor for VertexAI's client library.
+ An instrumentor for VertexAI client library.
"""
def instrumentation_dependencies(self) -> Collection[str]:
return _instruments
def _instrument(self, **kwargs):
- application_name = kwargs.get("application_name", "default")
+ version = importlib.metadata.version("google-cloud-aiplatform")
environment = kwargs.get("environment", "default")
+ application_name = kwargs.get("application_name", "default")
tracer = kwargs.get("tracer")
- metrics = kwargs.get("metrics_dict")
pricing_info = kwargs.get("pricing_info", {})
capture_message_content = kwargs.get("capture_message_content", False)
+ metrics = kwargs.get("metrics_dict")
disable_metrics = kwargs.get("disable_metrics")
- version = importlib.metadata.version("google-cloud-aiplatform")
- #sync
+ # sync generative models
wrap_function_wrapper(
"vertexai.generative_models",
"GenerativeModel.generate_content",
send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
wrap_function_wrapper(
"vertexai.generative_models",
"ChatSession.send_message",
send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
+ # sync language models
wrap_function_wrapper(
"vertexai.language_models",
"ChatSession.send_message",
send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
wrap_function_wrapper(
"vertexai.language_models",
"ChatSession.send_message_streaming",
send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
- #async
+ # async generative models
wrap_function_wrapper(
"vertexai.generative_models",
"GenerativeModel.generate_content_async",
async_send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
wrap_function_wrapper(
"vertexai.generative_models",
"ChatSession.send_message_async",
async_send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
+ # async language models
wrap_function_wrapper(
"vertexai.language_models",
"ChatSession.send_message_async",
async_send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
wrap_function_wrapper(
"vertexai.language_models",
"ChatSession.send_message_streaming_async",
async_send_message(version, environment, application_name,
- tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+ tracer, pricing_info, capture_message_content, metrics, disable_metrics),
)
def _uninstrument(self, **kwargs):
- # Proper uninstrumentation logic to revert patched methods
pass
diff --git a/sdk/python/src/openlit/instrumentation/vertexai/async_vertexai.py b/sdk/python/src/openlit/instrumentation/vertexai/async_vertexai.py
index ca2b05a9..9c2e4892 100644
--- a/sdk/python/src/openlit/instrumentation/vertexai/async_vertexai.py
+++ b/sdk/python/src/openlit/instrumentation/vertexai/async_vertexai.py
@@ -4,14 +4,15 @@ Module for monitoring VertexAI API calls.
import logging
import time
-from opentelemetry.trace import SpanKind, Status, StatusCode
-from opentelemetry.sdk.resources import SERVICE_NAME, TELEMETRY_SDK_NAME, DEPLOYMENT_ENVIRONMENT
+from opentelemetry.trace import SpanKind
from openlit.__helpers import (
- get_chat_model_cost,
handle_exception,
- calculate_ttft,
- calculate_tbt,
- create_metrics_attributes,
+)
+from openlit.instrumentation.vertexai.utils import (
+ process_chunk,
+ process_chat_response,
+ process_streaming_chat_response,
+ extract_vertexai_details,
)
from openlit.semcov import SemanticConvention
@@ -21,35 +22,19 @@ logger = logging.getLogger(__name__)
def async_send_message(version, environment, application_name, tracer,
pricing_info, capture_message_content, metrics, disable_metrics):
"""
- Generates a telemetry wrapper for messages to collect metrics.
-
- Args:
- version: Version of the monitoring package.
- environment: Deployment environment (e.g., production, staging).
- application_name: Name of the application using the VertexAI API.
- tracer: OpenTelemetry tracer for creating spans.
- pricing_info: Information used for calculating the cost of VertexAI usage.
- capture_message_content: Flag indicating whether to trace the actual content.
-
- Returns:
- A function that wraps the chat method to add telemetry.
+ Generates a telemetry wrapper for VertexAI AsyncMessages calls.
"""
class TracedAsyncStream:
"""
- Wrapper for streaming responses to collect metrics and trace data.
- Wraps the response to collect message IDs and aggregated response.
-
- This class implements the '__aiter__' and '__anext__' methods that
- handle asynchronous streaming responses.
-
- This class also implements '__aenter__' and '__aexit__' methods that
- handle asynchronous context management protocol.
+ Wrapper for async streaming responses to collect telemetry.
"""
+
def __init__(
self,
wrapped,
span,
+ span_name,
kwargs,
server_address,
server_port,
@@ -58,12 +43,10 @@ def async_send_message(version, environment, application_name, tracer,
):
self.__wrapped__ = wrapped
self._span = span
- # Placeholder for aggregating streaming response
+ self._span_name = span_name
self._llmresponse = ""
- self._input_tokens = ""
- self._output_tokens = ""
-
- self._args = args
+ self._input_tokens = 0
+ self._output_tokens = 0
self._kwargs = kwargs
self._start_time = time.time()
self._end_time = None
@@ -73,6 +56,7 @@ def async_send_message(version, environment, application_name, tracer,
self._server_address = server_address
self._server_port = server_port
self._request_model = request_model
+ self._args = args
async def __aenter__(self):
await self.__wrapped__.__aenter__()
@@ -91,369 +75,67 @@ def async_send_message(version, environment, application_name, tracer,
async def __anext__(self):
try:
chunk = await self.__wrapped__.__anext__()
- end_time = time.time()
- # Record the timestamp for the current chunk
- self._timestamps.append(end_time)
-
- if len(self._timestamps) == 1:
- # Calculate time to first chunk
- self._ttft = calculate_ttft(self._timestamps, self._start_time)
-
- self._llmresponse += str(chunk.text)
- self._input_tokens = chunk.usage_metadata.prompt_token_count
- self._output_tokens = chunk.usage_metadata.candidates_token_count
-
+ process_chunk(self, chunk)
return chunk
except StopAsyncIteration:
- # Handling exception ensure observability without disrupting operation
try:
- self._end_time = time.time()
- if len(self._timestamps) > 1:
- self._tbt = calculate_tbt(self._timestamps)
-
- # Format 'messages' into a single string
- message_prompt = self._kwargs.get("messages", "")
- formatted_messages = []
- for message in message_prompt:
- role = message["role"]
- content = message["content"]
-
- if isinstance(content, list):
- content_str_list = []
- for item in content:
- if item["type"] == "text":
- content_str_list.append(f'text: {item["text"]}')
- elif (item["type"] == "image_url" and
- not item["image_url"]["url"].startswith("data:")):
- content_str_list.append(f'image_url: {item["image_url"]["url"]}')
- content_str = ", ".join(content_str_list)
- formatted_messages.append(f"{role}: {content_str}")
- else:
- formatted_messages.append(f"{role}: {content}")
- prompt = "\n".join(formatted_messages) or str(self._args[0][0])
-
- # Calculate cost of the operation
- cost = get_chat_model_cost(self._request_model,
- pricing_info, self._input_tokens,
- self._output_tokens)
-
- # Set Span attributes (OTel Semconv)
- self._span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
- self._span.set_attribute(SemanticConvention.GEN_AI_OPERATION,
- SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT)
- self._span.set_attribute(SemanticConvention.GEN_AI_SYSTEM,
- SemanticConvention.GEN_AI_SYSTEM_VERTEXAI)
- self._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MODEL,
- self._request_model)
- self._span.set_attribute(SemanticConvention.SERVER_PORT,
- self._server_port)
-
- inference_config = self._kwargs.get('generation_config', {})
-
- # List of attributes and their config keys
- attributes = [
- (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, 'frequency_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, 'max_output_tokens'),
- (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, 'presence_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, 'stop_sequences'),
- (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, 'temperature'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_P, 'top_p'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_K, 'top_k'),
- ]
-
- # Set each attribute if the corresponding value exists and is not None
- for attribute, key in attributes:
- # Use the `get` method to safely access keys in the dictionary
- value = inference_config.get(key)
- if value is not None:
- self._span.set_attribute(attribute, value)
-
- self._span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_MODEL,
- self._request_model)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS,
- self._input_tokens)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS,
- self._output_tokens)
- self._span.set_attribute(SemanticConvention.SERVER_ADDRESS,
- self._server_address)
- if isinstance(self._llmresponse, str):
- self._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "text")
- else:
- self._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "json")
-
- # Set Span attributes (Extra)
- self._span.set_attribute(DEPLOYMENT_ENVIRONMENT,
- environment)
- self._span.set_attribute(SERVICE_NAME,
- application_name)
- self._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_IS_STREAM,
- True)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_TOTAL_TOKENS,
- self._input_tokens + self._output_tokens)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST,
- cost)
- self._span.set_attribute(SemanticConvention.GEN_AI_SERVER_TBT,
- self._tbt)
- self._span.set_attribute(SemanticConvention.GEN_AI_SERVER_TTFT,
- self._ttft)
- self._span.set_attribute(SemanticConvention.GEN_AI_SDK_VERSION,
- version)
- if capture_message_content:
- self._span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_PROMPT: prompt,
- },
- )
- self._span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_COMPLETION: self._llmresponse,
- },
- )
- self._span.set_status(Status(StatusCode.OK))
-
- if disable_metrics is False:
- attributes = create_metrics_attributes(
- service_name=application_name,
- deployment_environment=environment,
- operation=SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT,
- system=SemanticConvention.GEN_AI_SYSTEM_VERTEXAI,
- request_model=self._request_model,
- server_address=self._server_address,
- server_port=self._server_port,
- response_model=self._request_model,
- )
-
- metrics["genai_client_usage_tokens"].record(
- self._input_tokens + self._output_tokens, attributes
- )
- metrics["genai_client_operation_duration"].record(
- self._end_time - self._start_time, attributes
- )
- metrics["genai_server_tbt"].record(
- self._tbt, attributes
- )
- metrics["genai_server_ttft"].record(
- self._ttft, attributes
+ with self._span:
+ process_streaming_chat_response(
+ self,
+ pricing_info=pricing_info,
+ environment=environment,
+ application_name=application_name,
+ metrics=metrics,
+ capture_message_content=capture_message_content,
+ disable_metrics=disable_metrics,
+ version=version
)
- metrics["genai_requests"].add(1, attributes)
- metrics["genai_completion_tokens"].add(self._output_tokens, attributes)
- metrics["genai_prompt_tokens"].add(self._input_tokens, attributes)
- metrics["genai_cost"].record(cost, attributes)
-
except Exception as e:
handle_exception(self._span, e)
- logger.error("Error in trace creation: %s", e)
- finally:
- self._span.end()
raise
async def wrapper(wrapped, instance, args, kwargs):
"""
- Wraps the 'messages' API call to add telemetry.
-
- This collects metrics such as execution time, cost, and token usage, and handles errors
- gracefully, adding details to the trace for observability.
-
- Args:
- wrapped: The original 'messages' method to be wrapped.
- instance: The instance of the class where the original method is defined.
- args: Positional arguments for the 'messages' method.
- kwargs: Keyword arguments for the 'messages' method.
-
- Returns:
- The response from the original 'messages' method.
+ Wraps the VertexAI async API call to add telemetry.
"""
- # Check if streaming is enabled for the API call
streaming = kwargs.get("stream", False)
-
- try:
- location = instance._model._location
- request_model = "/".join(instance._model._model_name.split("/")[3:])
- except:
- location = instance._location
- request_model = "/".join(instance._model_name.split("/")[3:])
-
- server_address, server_port = location + '-aiplatform.googleapis.com', 443
+ server_address, server_port, request_model = extract_vertexai_details(instance)
span_name = f"{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}"
- # pylint: disable=no-else-return
if streaming:
- # Special handling for streaming response to accommodate the nature of data flow
awaited_wrapped = await wrapped(*args, **kwargs)
span = tracer.start_span(span_name, kind=SpanKind.CLIENT)
- return TracedAsyncStream(awaited_wrapped, span, kwargs, server_address, server_port, request_model, args)
+ return TracedAsyncStream(awaited_wrapped, span, span_name, kwargs, server_address, server_port, request_model, args)
- # Handling for non-streaming responses
else:
with tracer.start_as_current_span(span_name, kind=SpanKind.CLIENT) as span:
start_time = time.time()
response = await wrapped(*args, **kwargs)
- end_time = time.time()
try:
- # Format 'messages' into a single string
- message_prompt = kwargs.get("contents", [])
- formatted_messages = []
-
- for content in message_prompt:
- role = content.role
- parts = content.parts
- content_str = []
-
- for part in parts:
- # Collect relevant fields and handle each type of data that Part could contain
- if part.text:
- content_str.append(f"text: {part.text}")
- if part.video_metadata:
- content_str.append(f"video_metadata: {part.video_metadata}")
- if part.thought:
- content_str.append(f"thought: {part.thought}")
- if part.code_execution_result:
- content_str.append(f"code_execution_result: {part.code_execution_result}")
- if part.executable_code:
- content_str.append(f"executable_code: {part.executable_code}")
- if part.file_data:
- content_str.append(f"file_data: {part.file_data}")
- if part.function_call:
- content_str.append(f"function_call: {part.function_call}")
- if part.function_response:
- content_str.append(f"function_response: {part.function_response}")
- if part.inline_data:
- content_str.append(f"inline_data: {part.inline_data}")
-
- formatted_messages.append(f"{role}: {', '.join(content_str)}")
-
- prompt = "\n".join(formatted_messages) or str(args[0][0])
-
- input_tokens = response.usage_metadata.prompt_token_count
- output_tokens = response.usage_metadata.candidates_token_count
-
- # Calculate cost of the operation
- cost = get_chat_model_cost(request_model,
- pricing_info, input_tokens,
- output_tokens)
-
- # Set base span attribues (OTel Semconv)
- span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
- span.set_attribute(SemanticConvention.GEN_AI_OPERATION,
- SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT)
- span.set_attribute(SemanticConvention.GEN_AI_SYSTEM,
- SemanticConvention.GEN_AI_SYSTEM_VERTEXAI)
- span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MODEL,
- request_model)
- span.set_attribute(SemanticConvention.SERVER_PORT,
- server_port)
-
- inference_config = kwargs.get('generation_config', {})
-
- # List of attributes and their config keys
- attributes = [
- (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, 'frequency_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, 'max_output_tokens'),
- (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, 'presence_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, 'stop_sequences'),
- (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, 'temperature'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_P, 'top_p'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_K, 'top_k'),
- ]
-
- # Set each attribute if the corresponding value exists and is not None
- for attribute, key in attributes:
- # Use the `get` method to safely access keys in the dictionary
- value = inference_config.get(key)
- if value is not None:
- span.set_attribute(attribute, value)
-
- span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_MODEL,
- request_model)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS,
- input_tokens)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS,
- output_tokens)
- span.set_attribute(SemanticConvention.SERVER_ADDRESS,
- server_address)
- # span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_FINISH_REASON,
- # [str(response.candidates[0].finish_reason)])
-
- # Set base span attribues (Extras)
- span.set_attribute(DEPLOYMENT_ENVIRONMENT,
- environment)
- span.set_attribute(SERVICE_NAME,
- application_name)
- span.set_attribute(SemanticConvention.GEN_AI_REQUEST_IS_STREAM,
- False)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_TOTAL_TOKENS,
- input_tokens + output_tokens)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST,
- cost)
- span.set_attribute(SemanticConvention.GEN_AI_SERVER_TTFT,
- end_time - start_time)
- span.set_attribute(SemanticConvention.GEN_AI_SDK_VERSION,
- version)
- if capture_message_content:
- span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_PROMPT: prompt,
- },
- )
- span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_COMPLETION: response.text,
- },
- )
-
- if isinstance(response.text, str):
- span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "text")
- elif response.text is not None:
- span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "json")
-
- span.set_status(Status(StatusCode.OK))
-
- if disable_metrics is False:
- attributes = create_metrics_attributes(
- service_name=application_name,
- deployment_environment=environment,
- operation=SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT,
- system=SemanticConvention.GEN_AI_SYSTEM_VERTEXAI,
- request_model=request_model,
- server_address=server_address,
- server_port=server_port,
- response_model=request_model,
- )
-
- metrics["genai_client_usage_tokens"].record(
- input_tokens + output_tokens, attributes
- )
- metrics["genai_client_operation_duration"].record(
- end_time - start_time, attributes
- )
- metrics["genai_server_ttft"].record(
- end_time - start_time, attributes
- )
- metrics["genai_requests"].add(1, attributes)
- metrics["genai_completion_tokens"].add(output_tokens, attributes)
- metrics["genai_prompt_tokens"].add(input_tokens, attributes)
- metrics["genai_cost"].record(cost, attributes)
-
- # Return original response
- return response
+ response = process_chat_response(
+ response=response,
+ request_model=request_model,
+ pricing_info=pricing_info,
+ server_port=server_port,
+ server_address=server_address,
+ environment=environment,
+ application_name=application_name,
+ metrics=metrics,
+ start_time=start_time,
+ span=span,
+ capture_message_content=capture_message_content,
+ disable_metrics=disable_metrics,
+ version=version,
+ **kwargs
+ )
except Exception as e:
handle_exception(span, e)
- logger.error("Error in trace creation: %s", e)
- # Return original response
- return response
+ return response
return wrapper
diff --git a/sdk/python/src/openlit/instrumentation/vertexai/utils.py b/sdk/python/src/openlit/instrumentation/vertexai/utils.py
new file mode 100644
index 00000000..de1328b6
--- /dev/null
+++ b/sdk/python/src/openlit/instrumentation/vertexai/utils.py
@@ -0,0 +1,204 @@
+"""
+VertexAI OpenTelemetry instrumentation utility functions
+"""
+import time
+
+from opentelemetry.trace import Status, StatusCode
+
+from openlit.__helpers import (
+ calculate_ttft,
+ calculate_tbt,
+ get_chat_model_cost,
+ record_completion_metrics,
+ common_span_attributes,
+)
+from openlit.semcov import SemanticConvention
+
+def format_content(contents):
+ """
+ Format the VertexAI contents into a string for span events.
+ """
+
+ if not contents:
+ return ""
+
+ formatted_messages = []
+ for content in contents:
+ role = content.role
+ parts = content.parts
+ content_str = []
+
+ for part in parts:
+ # Collect relevant fields and handle each type of data that Part could contain
+ if part.text:
+ content_str.append(f"text: {part.text}")
+ if part.video_metadata:
+ content_str.append(f"video_metadata: {part.video_metadata}")
+ if part.thought:
+ content_str.append(f"thought: {part.thought}")
+ if part.code_execution_result:
+ content_str.append(f"code_execution_result: {part.code_execution_result}")
+ if part.executable_code:
+ content_str.append(f"executable_code: {part.executable_code}")
+ if part.file_data:
+ content_str.append(f"file_data: {part.file_data}")
+ if part.function_call:
+ content_str.append(f"function_call: {part.function_call}")
+ if part.function_response:
+ content_str.append(f"function_response: {part.function_response}")
+ if part.inline_data:
+ content_str.append(f"inline_data: {part.inline_data}")
+
+ formatted_messages.append(f"{role}: {', '.join(content_str)}")
+
+ return "\n".join(formatted_messages)
+
+def process_chunk(scope, chunk):
+ """
+ Process a chunk of response data and update state.
+ """
+
+ end_time = time.time()
+ # Record the timestamp for the current chunk
+ scope._timestamps.append(end_time)
+
+ if len(scope._timestamps) == 1:
+ # Calculate time to first chunk
+ scope._ttft = calculate_ttft(scope._timestamps, scope._start_time)
+
+ # Aggregate response content
+ scope._llmresponse += str(chunk.text)
+ scope._input_tokens = chunk.usage_metadata.prompt_token_count
+ scope._output_tokens = chunk.usage_metadata.candidates_token_count
+
+def common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+ capture_message_content, disable_metrics, version, is_stream):
+ """
+ Process chat request and generate Telemetry
+ """
+
+ scope._end_time = time.time()
+ if len(scope._timestamps) > 1:
+ scope._tbt = calculate_tbt(scope._timestamps)
+
+ # Format content using VertexAI-specific logic
+ contents = scope._kwargs.get("contents", [])
+ formatted_messages = format_content(contents)
+ prompt = formatted_messages or str(scope._args[0][0])
+
+ cost = get_chat_model_cost(scope._request_model, pricing_info, scope._input_tokens, scope._output_tokens)
+
+ # Common Span Attributes
+ common_span_attributes(scope,
+ SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_VERTEXAI,
+ scope._server_address, scope._server_port, scope._request_model, scope._request_model,
+ environment, application_name, is_stream, scope._tbt, scope._ttft, version)
+
+ # Span Attributes for Request parameters (VertexAI-specific)
+ inference_config = scope._kwargs.get("generation_config", {})
+
+ # List of attributes and their config keys
+ attributes = [
+ (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, "frequency_penalty"),
+ (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, "max_output_tokens"),
+ (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, "presence_penalty"),
+ (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, "stop_sequences"),
+ (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, "temperature"),
+ (SemanticConvention.GEN_AI_REQUEST_TOP_P, "top_p"),
+ (SemanticConvention.GEN_AI_REQUEST_TOP_K, "top_k"),
+ ]
+
+ # Set each attribute if the corresponding value exists and is not None
+ for attribute, key in attributes:
+ value = inference_config.get(key)
+ if value is not None:
+ scope._span.set_attribute(attribute, value)
+
+ # Span Attributes for Response parameters
+ scope._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE, "text" if isinstance(scope._llmresponse, str) else "json")
+
+ # Span Attributes for Cost and Tokens
+ scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS, scope._input_tokens)
+ scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS, scope._output_tokens)
+ scope._span.set_attribute(SemanticConvention.GEN_AI_CLIENT_TOKEN_USAGE, scope._input_tokens + scope._output_tokens)
+ scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST, cost)
+
+ # Span Attributes for Content
+ if capture_message_content:
+ scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_PROMPT, prompt)
+ scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_COMPLETION, scope._llmresponse)
+
+ # To be removed once the change to span_attributes (from span events) is complete
+ scope._span.add_event(
+ name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
+ attributes={
+ SemanticConvention.GEN_AI_CONTENT_PROMPT: prompt,
+ },
+ )
+ scope._span.add_event(
+ name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
+ attributes={
+ SemanticConvention.GEN_AI_CONTENT_COMPLETION: scope._llmresponse,
+ },
+ )
+
+ scope._span.set_status(Status(StatusCode.OK))
+
+ # Record metrics
+ if not disable_metrics:
+ record_completion_metrics(metrics, SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_VERTEXAI,
+ scope._server_address, scope._server_port, scope._request_model, scope._request_model, environment,
+ application_name, scope._start_time, scope._end_time, scope._input_tokens, scope._output_tokens,
+ cost, scope._tbt, scope._ttft)
+
+def process_streaming_chat_response(scope, pricing_info, environment, application_name, metrics,
+ capture_message_content=False, disable_metrics=False, version=""):
+ """
+ Process streaming chat response and generate telemetry.
+ """
+
+ common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+ capture_message_content, disable_metrics, version, is_stream=True)
+
+def process_chat_response(response, request_model, pricing_info, server_port, server_address,
+ environment, application_name, metrics, start_time,
+ span, capture_message_content=False, disable_metrics=False, version="1.0.0", **kwargs):
+ """
+ Process non-streaming chat response and generate telemetry.
+ """
+
+ scope = type("GenericScope", (), {})()
+
+ scope._start_time = start_time
+ scope._end_time = time.time()
+ scope._span = span
+ scope._llmresponse = response.text
+ scope._input_tokens = response.usage_metadata.prompt_token_count
+ scope._output_tokens = response.usage_metadata.candidates_token_count
+ scope._timestamps = []
+ scope._ttft, scope._tbt = scope._end_time - scope._start_time, 0
+ scope._server_address, scope._server_port = server_address, server_port
+ scope._request_model = request_model
+ scope._kwargs = kwargs
+ scope._args = [kwargs.get("contents", [])]
+
+ common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+ capture_message_content, disable_metrics, version, is_stream=False)
+
+ return response
+
+def extract_vertexai_details(instance):
+ """
+ Extract VertexAI-specific details like location and model name.
+ """
+ try:
+ location = instance._model._location
+ request_model = "/".join(instance._model._model_name.split("/")[3:])
+ except:
+ location = instance._location
+ request_model = "/".join(instance._model_name.split("/")[3:])
+
+ server_address = location + "-aiplatform.googleapis.com"
+ server_port = 443
+
+ return server_address, server_port, request_model
diff --git a/sdk/python/src/openlit/instrumentation/vertexai/vertexai.py b/sdk/python/src/openlit/instrumentation/vertexai/vertexai.py
index 811c6729..905ca68d 100644
--- a/sdk/python/src/openlit/instrumentation/vertexai/vertexai.py
+++ b/sdk/python/src/openlit/instrumentation/vertexai/vertexai.py
@@ -4,14 +4,15 @@ Module for monitoring VertexAI API calls.
import logging
import time
-from opentelemetry.trace import SpanKind, Status, StatusCode
-from opentelemetry.sdk.resources import SERVICE_NAME, TELEMETRY_SDK_NAME, DEPLOYMENT_ENVIRONMENT
+from opentelemetry.trace import SpanKind
from openlit.__helpers import (
- get_chat_model_cost,
handle_exception,
- calculate_ttft,
- calculate_tbt,
- create_metrics_attributes,
+)
+from openlit.instrumentation.vertexai.utils import (
+ process_chunk,
+ process_chat_response,
+ process_streaming_chat_response,
+ extract_vertexai_details,
)
from openlit.semcov import SemanticConvention
@@ -21,35 +22,19 @@ logger = logging.getLogger(__name__)
def send_message(version, environment, application_name, tracer,
pricing_info, capture_message_content, metrics, disable_metrics):
"""
- Generates a telemetry wrapper for messages to collect metrics.
-
- Args:
- version: Version of the monitoring package.
- environment: Deployment environment (e.g., production, staging).
- application_name: Name of the application using the VertexAI API.
- tracer: OpenTelemetry tracer for creating spans.
- pricing_info: Information used for calculating the cost of VertexAI usage.
- capture_message_content: Flag indicating whether to trace the actual content.
-
- Returns:
- A function that wraps the chat method to add telemetry.
+ Generates a telemetry wrapper for VertexAI messages to collect metrics.
"""
class TracedSyncStream:
"""
- Wrapper for streaming responses to collect metrics and trace data.
- Wraps the response to collect message IDs and aggregated response.
-
- This class implements the '__aiter__' and '__anext__' methods that
- handle asynchronous streaming responses.
-
- This class also implements '__aenter__' and '__aexit__' methods that
- handle asynchronous context management protocol.
+ Wrapper for streaming responses to collect telemetry.
"""
+
def __init__(
self,
wrapped,
span,
+ span_name,
kwargs,
server_address,
server_port,
@@ -58,12 +43,10 @@ def send_message(version, environment, application_name, tracer,
):
self.__wrapped__ = wrapped
self._span = span
- # Placeholder for aggregating streaming response
+ self._span_name = span_name
self._llmresponse = ""
- self._input_tokens = ""
- self._output_tokens = ""
-
- self._args = args
+ self._input_tokens = 0
+ self._output_tokens = 0
self._kwargs = kwargs
self._start_time = time.time()
self._end_time = None
@@ -73,6 +56,7 @@ def send_message(version, environment, application_name, tracer,
self._server_address = server_address
self._server_port = server_port
self._request_model = request_model
+ self._args = args
def __enter__(self):
self.__wrapped__.__enter__()
@@ -91,369 +75,67 @@ def send_message(version, environment, application_name, tracer,
def __next__(self):
try:
chunk = self.__wrapped__.__next__()
- end_time = time.time()
- # Record the timestamp for the current chunk
- self._timestamps.append(end_time)
-
- if len(self._timestamps) == 1:
- # Calculate time to first chunk
- self._ttft = calculate_ttft(self._timestamps, self._start_time)
-
- self._llmresponse += str(chunk.text)
- self._input_tokens = chunk.usage_metadata.prompt_token_count
- self._output_tokens = chunk.usage_metadata.candidates_token_count
-
+ process_chunk(self, chunk)
return chunk
except StopIteration:
- # Handling exception ensure observability without disrupting operation
try:
- self._end_time = time.time()
- if len(self._timestamps) > 1:
- self._tbt = calculate_tbt(self._timestamps)
-
- # Format 'messages' into a single string
- message_prompt = self._kwargs.get("messages", "")
- formatted_messages = []
- for message in message_prompt:
- role = message["role"]
- content = message["content"]
-
- if isinstance(content, list):
- content_str_list = []
- for item in content:
- if item["type"] == "text":
- content_str_list.append(f'text: {item["text"]}')
- elif (item["type"] == "image_url" and
- not item["image_url"]["url"].startswith("data:")):
- content_str_list.append(f'image_url: {item["image_url"]["url"]}')
- content_str = ", ".join(content_str_list)
- formatted_messages.append(f"{role}: {content_str}")
- else:
- formatted_messages.append(f"{role}: {content}")
- prompt = "\n".join(formatted_messages) or str(self._args[0][0])
-
- # Calculate cost of the operation
- cost = get_chat_model_cost(self._request_model,
- pricing_info, self._input_tokens,
- self._output_tokens)
-
- # Set Span attributes (OTel Semconv)
- self._span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
- self._span.set_attribute(SemanticConvention.GEN_AI_OPERATION,
- SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT)
- self._span.set_attribute(SemanticConvention.GEN_AI_SYSTEM,
- SemanticConvention.GEN_AI_SYSTEM_VERTEXAI)
- self._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MODEL,
- self._request_model)
- self._span.set_attribute(SemanticConvention.SERVER_PORT,
- self._server_port)
-
- inference_config = self._kwargs.get('generation_config', {})
-
- # List of attributes and their config keys
- attributes = [
- (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, 'frequency_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, 'max_output_tokens'),
- (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, 'presence_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, 'stop_sequences'),
- (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, 'temperature'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_P, 'top_p'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_K, 'top_k'),
- ]
-
- # Set each attribute if the corresponding value exists and is not None
- for attribute, key in attributes:
- # Use the `get` method to safely access keys in the dictionary
- value = inference_config.get(key)
- if value is not None:
- self._span.set_attribute(attribute, value)
-
- self._span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_MODEL,
- self._request_model)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS,
- self._input_tokens)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS,
- self._output_tokens)
- self._span.set_attribute(SemanticConvention.SERVER_ADDRESS,
- self._server_address)
- if isinstance(self._llmresponse, str):
- self._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "text")
- else:
- self._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "json")
-
- # Set Span attributes (Extra)
- self._span.set_attribute(DEPLOYMENT_ENVIRONMENT,
- environment)
- self._span.set_attribute(SERVICE_NAME,
- application_name)
- self._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_IS_STREAM,
- True)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_TOTAL_TOKENS,
- self._input_tokens + self._output_tokens)
- self._span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST,
- cost)
- self._span.set_attribute(SemanticConvention.GEN_AI_SERVER_TBT,
- self._tbt)
- self._span.set_attribute(SemanticConvention.GEN_AI_SERVER_TTFT,
- self._ttft)
- self._span.set_attribute(SemanticConvention.GEN_AI_SDK_VERSION,
- version)
- if capture_message_content:
- self._span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_PROMPT: prompt,
- },
- )
- self._span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_COMPLETION: self._llmresponse,
- },
- )
- self._span.set_status(Status(StatusCode.OK))
-
- if disable_metrics is False:
- attributes = create_metrics_attributes(
- service_name=application_name,
- deployment_environment=environment,
- operation=SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT,
- system=SemanticConvention.GEN_AI_SYSTEM_VERTEXAI,
- request_model=self._request_model,
- server_address=self._server_address,
- server_port=self._server_port,
- response_model=self._request_model,
- )
-
- metrics["genai_client_usage_tokens"].record(
- self._input_tokens + self._output_tokens, attributes
- )
- metrics["genai_client_operation_duration"].record(
- self._end_time - self._start_time, attributes
- )
- metrics["genai_server_tbt"].record(
- self._tbt, attributes
- )
- metrics["genai_server_ttft"].record(
- self._ttft, attributes
+ with self._span:
+ process_streaming_chat_response(
+ self,
+ pricing_info=pricing_info,
+ environment=environment,
+ application_name=application_name,
+ metrics=metrics,
+ capture_message_content=capture_message_content,
+ disable_metrics=disable_metrics,
+ version=version
)
- metrics["genai_requests"].add(1, attributes)
- metrics["genai_completion_tokens"].add(self._output_tokens, attributes)
- metrics["genai_prompt_tokens"].add(self._input_tokens, attributes)
- metrics["genai_cost"].record(cost, attributes)
-
except Exception as e:
handle_exception(self._span, e)
- logger.error("Error in trace creation: %s", e)
- finally:
- self._span.end()
raise
def wrapper(wrapped, instance, args, kwargs):
"""
- Wraps the 'messages' API call to add telemetry.
-
- This collects metrics such as execution time, cost, and token usage, and handles errors
- gracefully, adding details to the trace for observability.
-
- Args:
- wrapped: The original 'messages' method to be wrapped.
- instance: The instance of the class where the original method is defined.
- args: Positional arguments for the 'messages' method.
- kwargs: Keyword arguments for the 'messages' method.
-
- Returns:
- The response from the original 'messages' method.
+ Wraps the VertexAI API call to add telemetry.
"""
- # Check if streaming is enabled for the API call
streaming = kwargs.get("stream", False)
-
- try:
- location = instance._model._location
- request_model = "/".join(instance._model._model_name.split("/")[3:])
- except:
- location = instance._location
- request_model = "/".join(instance._model_name.split("/")[3:])
-
- server_address, server_port = location + '-aiplatform.googleapis.com', 443
+ server_address, server_port, request_model = extract_vertexai_details(instance)
span_name = f"{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}"
- # pylint: disable=no-else-return
if streaming:
- # Special handling for streaming response to accommodate the nature of data flow
awaited_wrapped = wrapped(*args, **kwargs)
span = tracer.start_span(span_name, kind=SpanKind.CLIENT)
- return TracedSyncStream(awaited_wrapped, span, kwargs, server_address, server_port, request_model, args)
+ return TracedSyncStream(awaited_wrapped, span, span_name, kwargs, server_address, server_port, request_model, args)
- # Handling for non-streaming responses
else:
with tracer.start_as_current_span(span_name, kind=SpanKind.CLIENT) as span:
start_time = time.time()
response = wrapped(*args, **kwargs)
- end_time = time.time()
try:
- # Format 'messages' into a single string
- message_prompt = kwargs.get("contents", [])
- formatted_messages = []
-
- for content in message_prompt:
- role = content.role
- parts = content.parts
- content_str = []
-
- for part in parts:
- # Collect relevant fields and handle each type of data that Part could contain
- if part.text:
- content_str.append(f"text: {part.text}")
- if part.video_metadata:
- content_str.append(f"video_metadata: {part.video_metadata}")
- if part.thought:
- content_str.append(f"thought: {part.thought}")
- if part.code_execution_result:
- content_str.append(f"code_execution_result: {part.code_execution_result}")
- if part.executable_code:
- content_str.append(f"executable_code: {part.executable_code}")
- if part.file_data:
- content_str.append(f"file_data: {part.file_data}")
- if part.function_call:
- content_str.append(f"function_call: {part.function_call}")
- if part.function_response:
- content_str.append(f"function_response: {part.function_response}")
- if part.inline_data:
- content_str.append(f"inline_data: {part.inline_data}")
-
- formatted_messages.append(f"{role}: {', '.join(content_str)}")
-
- prompt = "\n".join(formatted_messages) or str(args[0][0])
-
- input_tokens = response.usage_metadata.prompt_token_count
- output_tokens = response.usage_metadata.candidates_token_count
-
- # Calculate cost of the operation
- cost = get_chat_model_cost(request_model,
- pricing_info, input_tokens,
- output_tokens)
-
- # Set base span attribues (OTel Semconv)
- span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
- span.set_attribute(SemanticConvention.GEN_AI_OPERATION,
- SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT)
- span.set_attribute(SemanticConvention.GEN_AI_SYSTEM,
- SemanticConvention.GEN_AI_SYSTEM_VERTEXAI)
- span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MODEL,
- request_model)
- span.set_attribute(SemanticConvention.SERVER_PORT,
- server_port)
-
- inference_config = kwargs.get('generation_config', {})
-
- # List of attributes and their config keys
- attributes = [
- (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, 'frequency_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, 'max_output_tokens'),
- (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, 'presence_penalty'),
- (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, 'stop_sequences'),
- (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, 'temperature'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_P, 'top_p'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_K, 'top_k'),
- ]
-
- # Set each attribute if the corresponding value exists and is not None
- for attribute, key in attributes:
- # Use the `get` method to safely access keys in the dictionary
- value = inference_config.get(key)
- if value is not None:
- span.set_attribute(attribute, value)
-
- span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_MODEL,
- request_model)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS,
- input_tokens)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS,
- output_tokens)
- span.set_attribute(SemanticConvention.SERVER_ADDRESS,
- server_address)
- # span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_FINISH_REASON,
- # [str(response.candidates[0].finish_reason)])
-
- # Set base span attribues (Extras)
- span.set_attribute(DEPLOYMENT_ENVIRONMENT,
- environment)
- span.set_attribute(SERVICE_NAME,
- application_name)
- span.set_attribute(SemanticConvention.GEN_AI_REQUEST_IS_STREAM,
- False)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_TOTAL_TOKENS,
- input_tokens + output_tokens)
- span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST,
- cost)
- span.set_attribute(SemanticConvention.GEN_AI_SERVER_TTFT,
- end_time - start_time)
- span.set_attribute(SemanticConvention.GEN_AI_SDK_VERSION,
- version)
- if capture_message_content:
- span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_PROMPT: prompt,
- },
- )
- span.add_event(
- name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
- attributes={
- SemanticConvention.GEN_AI_CONTENT_COMPLETION: response.text,
- },
- )
-
- if isinstance(response.text, str):
- span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "text")
- elif response.text is not None:
- span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- "json")
-
- span.set_status(Status(StatusCode.OK))
-
- if disable_metrics is False:
- attributes = create_metrics_attributes(
- service_name=application_name,
- deployment_environment=environment,
- operation=SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT,
- system=SemanticConvention.GEN_AI_SYSTEM_VERTEXAI,
- request_model=request_model,
- server_address=server_address,
- server_port=server_port,
- response_model=request_model,
- )
-
- metrics["genai_client_usage_tokens"].record(
- input_tokens + output_tokens, attributes
- )
- metrics["genai_client_operation_duration"].record(
- end_time - start_time, attributes
- )
- metrics["genai_server_ttft"].record(
- end_time - start_time, attributes
- )
- metrics["genai_requests"].add(1, attributes)
- metrics["genai_completion_tokens"].add(output_tokens, attributes)
- metrics["genai_prompt_tokens"].add(input_tokens, attributes)
- metrics["genai_cost"].record(cost, attributes)
-
- # Return original response
- return response
+ response = process_chat_response(
+ response=response,
+ request_model=request_model,
+ pricing_info=pricing_info,
+ server_port=server_port,
+ server_address=server_address,
+ environment=environment,
+ application_name=application_name,
+ metrics=metrics,
+ start_time=start_time,
+ span=span,
+ capture_message_content=capture_message_content,
+ disable_metrics=disable_metrics,
+ version=version,
+ **kwargs
+ )
except Exception as e:
handle_exception(span, e)
- logger.error("Error in trace creation: %s", e)
- # Return original response
- return response
+ return response
return wrapper