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 AWS Bedrock OpenTelemetry Monitoring Instrumentation (#783)

Patcher9fa971cf57e3
diff --git a/sdk/python/pyproject.toml b/sdk/python/pyproject.toml
index d8f1ece2..4dde472b 100644
--- a/sdk/python/pyproject.toml
+++ b/sdk/python/pyproject.toml
@@ -1,6 +1,6 @@
[tool.poetry]
name = "openlit"
-version = "1.34.19"
+version = "1.34.20"
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/bedrock/__init__.py b/sdk/python/src/openlit/instrumentation/bedrock/__init__.py
index 89a32efc..23aa5a22 100644
--- a/sdk/python/src/openlit/instrumentation/bedrock/__init__.py
+++ b/sdk/python/src/openlit/instrumentation/bedrock/__init__.py
@@ -1,4 +1,3 @@
-# pylint: disable=useless-return, bad-staticmethod-argument, disable=duplicate-code
"""Initializer of Auto Instrumentation of AWS Bedrock Functions"""
from typing import Collection
@@ -6,37 +5,43 @@ import importlib.metadata
from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
from wrapt import wrap_function_wrapper
-from openlit.instrumentation.bedrock.bedrock import converse
+from openlit.instrumentation.bedrock.bedrock import converse, converse_stream
_instruments = ("boto3 >= 1.34.138",)
class BedrockInstrumentor(BaseInstrumentor):
"""
- An instrumentor for AWS Bedrock's client library.
+ An instrumentor for AWS Bedrock client library.
"""
def instrumentation_dependencies(self) -> Collection[str]:
return _instruments
def _instrument(self, **kwargs):
- application_name = kwargs.get("application_name", "default_application")
- environment = kwargs.get("environment", "default_environment")
+ version = importlib.metadata.version("boto3")
+ environment = kwargs.get("environment", "default")
+ application_name = kwargs.get("application_name", "default")
tracer = kwargs.get("tracer")
- event_provider = kwargs.get('event_provider')
- 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("boto3")
- #sync
+ # sync
+ wrap_function_wrapper(
+ "botocore.client",
+ "ClientCreator.create_client",
+ converse(version, environment, application_name, tracer, pricing_info,
+ capture_message_content, metrics, disable_metrics),
+ )
+
+ # streaming
wrap_function_wrapper(
- "botocore.client",
- "ClientCreator.create_client",
- converse(version, environment, application_name,
- tracer, event_provider, pricing_info, capture_message_content, metrics, disable_metrics),
+ "botocore.client",
+ "ClientCreator.create_client",
+ converse_stream(version, environment, application_name, 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/bedrock/bedrock.py b/sdk/python/src/openlit/instrumentation/bedrock/bedrock.py
index d1804cf9..b617e325 100644
--- a/sdk/python/src/openlit/instrumentation/bedrock/bedrock.py
+++ b/sdk/python/src/openlit/instrumentation/bedrock/bedrock.py
@@ -2,64 +2,65 @@
Module for monitoring Amazon Bedrock API calls.
"""
-import logging
import time
from opentelemetry.trace import SpanKind
from openlit.__helpers import (
+ handle_exception,
set_server_address_and_port
)
from openlit.instrumentation.bedrock.utils import (
+ process_chunk,
process_chat_response,
+ process_streaming_chat_response,
)
from openlit.semcov import SemanticConvention
-# Initialize logger for logging potential issues and operations
-logger = logging.getLogger(__name__)
-
-def converse(version, environment, application_name, tracer, event_provider,
- pricing_info, capture_message_content, metrics, disable_metrics):
+def converse(version, environment, application_name, tracer, pricing_info, capture_message_content, metrics, disable_metrics):
"""
- Generates a telemetry wrapper for GenAI function call
+ Generates a telemetry wrapper for AWS Bedrock converse calls.
"""
def wrapper(wrapped, instance, args, kwargs):
"""
- Wraps the GenAI function call.
+ Wraps the ClientCreator.create_client call.
"""
def converse_wrapper(original_method, *method_args, **method_kwargs):
-
"""
- Wraps the GenAI function call.
+ Wraps the individual converse method call.
"""
- server_address, server_port = set_server_address_and_port(instance, 'aws.amazon.com', 443)
- request_model = method_kwargs.get('modelId', 'amazon.titan-text-express-v1')
+ server_address, server_port = set_server_address_and_port(instance, "aws.amazon.com", 443)
+ request_model = method_kwargs.get("modelId", "amazon.titan-text-express-v1")
- span_name = f'{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}'
+ span_name = f"{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}"
with tracer.start_as_current_span(span_name, kind=SpanKind.CLIENT) as span:
start_time = time.time()
response = original_method(*method_args, **method_kwargs)
- llm_config = method_kwargs.get('inferenceConfig', {})
- 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,
- event_provider=event_provider,
- start_time=start_time,
- span=span,
- capture_message_content=capture_message_content,
- disable_metrics=disable_metrics,
- version=version,
- llm_config=llm_config,
- **method_kwargs
- )
+ llm_config = method_kwargs.get("inferenceConfig", {})
+
+ try:
+ 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,
+ llm_config=llm_config,
+ **method_kwargs
+ )
+
+ except Exception as e:
+ handle_exception(span, e)
return response
@@ -67,10 +68,143 @@ def converse(version, environment, application_name, tracer, event_provider,
client = wrapped(*args, **kwargs)
# Replace the original method with the instrumented one
- if kwargs.get('service_name') == 'bedrock-runtime':
+ if kwargs.get("service_name") == "bedrock-runtime":
original_invoke_model = client.converse
- client.converse = lambda *args, **kwargs: converse_wrapper(original_invoke_model,
- *args, **kwargs)
+ client.converse = lambda *args, **kwargs: converse_wrapper(original_invoke_model, *args, **kwargs)
+
+ return client
+
+ return wrapper
+
+def converse_stream(version, environment, application_name, tracer, pricing_info, capture_message_content, metrics, disable_metrics):
+ """
+ Generates a telemetry wrapper for AWS Bedrock converse_stream calls.
+ """
+
+ class TracedSyncStream:
+ """
+ Wrapper for streaming responses to collect telemetry.
+ """
+
+ def __init__(
+ self,
+ wrapped_response,
+ span,
+ span_name,
+ kwargs,
+ server_address,
+ server_port,
+ **args,
+ ):
+ self.__wrapped_response = wrapped_response
+ # Extract the actual stream iterator from the response
+ if isinstance(wrapped_response, dict) and "stream" in wrapped_response:
+ self.__wrapped_stream = iter(wrapped_response["stream"])
+ else:
+ self.__wrapped_stream = iter(wrapped_response)
+
+ self._span = span
+ self._span_name = span_name
+ self._llmresponse = ""
+ self._response_id = ""
+ self._response_model = ""
+ self._finish_reason = ""
+ self._tools = None
+ self._input_tokens = 0
+ self._output_tokens = 0
+
+ self._args = args
+ self._kwargs = kwargs
+ self._start_time = time.time()
+ self._end_time = None
+ self._timestamps = []
+ self._ttft = 0
+ self._tbt = 0
+ self._server_address = server_address
+ self._server_port = server_port
+
+ def __enter__(self):
+ if hasattr(self.__wrapped_stream, "__enter__"):
+ self.__wrapped_stream.__enter__()
+ return self
+
+ def __exit__(self, exc_type, exc_value, traceback):
+ if hasattr(self.__wrapped_stream, "__exit__"):
+ self.__wrapped_stream.__exit__(exc_type, exc_value, traceback)
+
+ def __iter__(self):
+ return self
+
+ def __getattr__(self, name):
+ """Delegate attribute access to the wrapped response."""
+ return getattr(self.__wrapped_response, name)
+
+ def get(self, key, default=None):
+ """Delegate get method to the wrapped response if its a dict."""
+ if isinstance(self.__wrapped_response, dict):
+ return self.__wrapped_response.get(key, default)
+ return getattr(self.__wrapped_response, key, default)
+
+ def __getitem__(self, key):
+ """Delegate item access to the wrapped response if its a dict."""
+ if isinstance(self.__wrapped_response, dict):
+ return self.__wrapped_response[key]
+ return getattr(self.__wrapped_response, key)
+
+ def __next__(self):
+ try:
+ chunk = next(self.__wrapped_stream)
+ process_chunk(self, chunk)
+ return chunk
+ except StopIteration:
+ try:
+ llm_config = self._kwargs.get("inferenceConfig", {})
+ with tracer.start_as_current_span(self._span_name, kind=SpanKind.CLIENT) as 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,
+ llm_config=llm_config
+ )
+
+ except Exception as e:
+ handle_exception(self._span, e)
+
+ raise
+
+ def wrapper(wrapped, instance, args, kwargs):
+ """
+ Wraps the ClientCreator.create_client call.
+ """
+
+ def converse_stream_wrapper(original_method, *method_args, **method_kwargs):
+ """
+ Wraps the individual converse_stream method call.
+ """
+
+ server_address, server_port = set_server_address_and_port(instance, "aws.amazon.com", 443)
+ request_model = method_kwargs.get("modelId", "amazon.titan-text-express-v1")
+
+ span_name = f"{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}"
+
+ # Get the streaming response
+ stream_response = original_method(*method_args, **method_kwargs)
+ span = tracer.start_span(span_name, kind=SpanKind.CLIENT)
+
+ return TracedSyncStream(stream_response, span, span_name, method_kwargs, server_address, server_port)
+
+ # Get the original client instance from the wrapper
+ client = wrapped(*args, **kwargs)
+
+ # Replace the original method with the instrumented one
+ if kwargs.get("service_name") == "bedrock-runtime":
+ original_stream_model = client.converse_stream
+ client.converse_stream = lambda *args, **kwargs: converse_stream_wrapper(original_stream_model, *args, **kwargs)
return client
diff --git a/sdk/python/src/openlit/instrumentation/bedrock/utils.py b/sdk/python/src/openlit/instrumentation/bedrock/utils.py
index 949fa7bc..65aa429c 100644
--- a/sdk/python/src/openlit/instrumentation/bedrock/utils.py
+++ b/sdk/python/src/openlit/instrumentation/bedrock/utils.py
@@ -3,21 +3,55 @@ AWS Bedrock OpenTelemetry instrumentation utility functions
"""
import time
-from opentelemetry.sdk.resources import SERVICE_NAME, TELEMETRY_SDK_NAME, DEPLOYMENT_ENVIRONMENT
from opentelemetry.trace import Status, StatusCode
from openlit.__helpers import (
calculate_ttft,
response_as_dict,
calculate_tbt,
- extract_and_format_input,
get_chat_model_cost,
- create_metrics_attributes,
- otel_event,
- concatenate_all_contents
+ record_completion_metrics,
+ common_span_attributes,
+ handle_exception
)
from openlit.semcov import SemanticConvention
+def format_content(messages):
+ """
+ Format the messages into a string for span events.
+ """
+
+ if not messages:
+ return ""
+
+ formatted_messages = []
+ for message in messages:
+ if isinstance(message, dict):
+ role = message.get("role", "user")
+ content = message.get("content", "")
+ else:
+ # Handle Bedrock object format
+ role = getattr(message, "role", "user")
+ content = getattr(message, "content", "")
+
+ if isinstance(content, list):
+ # Handle structured content (e.g., text + images)
+ text_parts = []
+ for part in content:
+ if isinstance(part, dict):
+ # Bedrock format: {"text": "content"} or generic format: {"type": "text", "text": "content"}
+ if "text" in part:
+ text_parts.append(part.get("text", ""))
+ elif part.get("type") == "text":
+ text_parts.append(part.get("text", ""))
+ content = " ".join(text_parts)
+ elif not isinstance(content, str):
+ content = str(content)
+
+ formatted_messages.append(f"{role}: {content}")
+
+ return "\n".join(formatted_messages)
+
def process_chunk(self, chunk):
"""
Process a chunk of response data and update state.
@@ -33,33 +67,33 @@ def process_chunk(self, chunk):
chunked = response_as_dict(chunk)
- # Collect message IDs and input token from events
- if chunked.get('type') == 'message_start':
- self._response_id = chunked.get('message').get('id')
- self._input_tokens = chunked.get('message').get('usage').get('input_tokens')
- self._response_model = chunked.get('message').get('model')
- self._response_role = chunked.get('message').get('role')
-
- # Collect message IDs and aggregated response from events
- if chunked.get('type') == 'content_block_delta':
- if chunked.get('delta').get('text'):
- self._llmresponse += chunked.get('delta').get('text')
- elif chunked.get('delta').get('partial_json'):
- self._tool_arguments += chunked.get('delta').get('partial_json')
-
- if chunked.get('type') == 'content_block_start':
- if chunked.get('content_block').get('id'):
- self._tool_id = chunked.get('content_block').get('id')
- if chunked.get('content_block').get('name'):
- self._tool_name = chunked.get('content_block').get('name')
-
- # Collect output tokens and stop reason from events
- if chunked.get('type') == 'message_delta':
- self._output_tokens = chunked.get('usage').get('output_tokens')
- self._finish_reason = chunked.get('delta').get('stop_reason')
+ # Handle Bedrock messageStart event
+ if "messageStart" in chunked:
+ message_start = chunked.get("messageStart", {})
+ self._response_role = message_start.get("role", "assistant")
+
+ # Handle Bedrock contentBlockDelta event
+ if "contentBlockDelta" in chunked:
+ content_delta = chunked.get("contentBlockDelta", {})
+ delta = content_delta.get("delta", {})
+ if "text" in delta:
+ self._llmresponse += delta.get("text", "")
+
+ # Handle Bedrock messageStop event
+ if "messageStop" in chunked:
+ message_stop = chunked.get("messageStop", {})
+ self._finish_reason = message_stop.get("stopReason", "")
+
+ # Handle Bedrock metadata event (final event with usage info)
+ if "metadata" in chunked:
+ metadata = chunked.get("metadata", {})
+ usage = metadata.get("usage", {})
+ self._input_tokens = usage.get("inputTokens", 0)
+ self._output_tokens = usage.get("outputTokens", 0)
+ self._end_time = end_time
def common_chat_logic(scope, pricing_info, environment, application_name, metrics,
- event_provider, capture_message_content, disable_metrics, version, llm_config, is_stream):
+ capture_message_content, disable_metrics, version, llm_config, is_stream):
"""
Process chat request and generate Telemetry
"""
@@ -68,62 +102,55 @@ def common_chat_logic(scope, pricing_info, environment, application_name, metric
if len(scope._timestamps) > 1:
scope._tbt = calculate_tbt(scope._timestamps)
- formatted_messages = extract_and_format_input(scope._kwargs.get('messages', ''))
- print(formatted_messages)
- request_model = scope._kwargs.get('model', 'claude-3-opus-20240229')
+ formatted_messages = format_content(scope._kwargs.get("messages", []))
+ request_model = scope._kwargs.get("modelId", "amazon.titan-text-express-v1")
cost = get_chat_model_cost(request_model, pricing_info, scope._input_tokens, scope._output_tokens)
- # Set Span attributes (OTel Semconv)
- scope._span.set_attribute(TELEMETRY_SDK_NAME, 'openlit')
- scope._span.set_attribute(SemanticConvention.GEN_AI_OPERATION, SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT)
- scope._span.set_attribute(SemanticConvention.GEN_AI_SYSTEM, SemanticConvention.GEN_AI_SYSTEM_AWS_BEDROCK)
- scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MODEL, request_model)
- scope._span.set_attribute(SemanticConvention.SERVER_PORT, scope._server_port)
-
- # List of attributes and their config keys
- attributes = [
- (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, 'frequencyPenalty'),
- (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, 'maxTokens'),
- (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, 'presencePenalty'),
- (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, 'stopSequences'),
- (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, 'temperature'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_P, 'topP'),
- (SemanticConvention.GEN_AI_REQUEST_TOP_K, 'topK'),
+ # Common Span Attributes
+ common_span_attributes(scope,
+ SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_AWS_BEDROCK,
+ scope._server_address, scope._server_port, request_model, scope._response_model,
+ environment, application_name, is_stream, scope._tbt, scope._ttft, version)
+
+ # Bedrock-specific attributes from llm_config
+ bedrock_attributes = [
+ (SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY, "frequencyPenalty"),
+ (SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, "maxTokens"),
+ (SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY, "presencePenalty"),
+ (SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, "stopSequences"),
+ (SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, "temperature"),
+ (SemanticConvention.GEN_AI_REQUEST_TOP_P, "topP"),
+ (SemanticConvention.GEN_AI_REQUEST_TOP_K, "topK"),
]
- # Set each attribute if the corresponding value exists and is not None
- for attribute, key in attributes:
+ # Set each bedrock-specific attribute if the corresponding value exists and is not None
+ for attribute, key in bedrock_attributes:
value = llm_config.get(key)
if value is not None:
scope._span.set_attribute(attribute, value)
- scope._span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_FINISH_REASON, [scope._finish_reason])
+ # Span Attributes for Response parameters
scope._span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_ID, scope._response_id)
- scope._span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_MODEL, scope._response_model)
+ scope._span.set_attribute(SemanticConvention.GEN_AI_RESPONSE_FINISH_REASON, [scope._finish_reason])
+ 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.SERVER_ADDRESS, scope._server_address)
-
- scope._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE,
- 'text' if isinstance(scope._llmresponse, str) else 'json')
-
- scope._span.set_attribute(DEPLOYMENT_ENVIRONMENT, environment)
- scope._span.set_attribute(SERVICE_NAME, application_name)
- scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_IS_STREAM, is_stream)
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)
- scope._span.set_attribute(SemanticConvention.GEN_AI_SERVER_TBT, scope._tbt)
- scope._span.set_attribute(SemanticConvention.GEN_AI_SERVER_TTFT, scope._ttft)
- scope._span.set_attribute(SemanticConvention.GEN_AI_SDK_VERSION, version)
- # To be removed one the change to log events (from span events) is complete
- prompt = concatenate_all_contents(formatted_messages)
+ # Span Attributes for Content
if capture_message_content:
+ scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_PROMPT, formatted_messages)
+ 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,
+ SemanticConvention.GEN_AI_CONTENT_PROMPT: formatted_messages,
},
)
scope._span.add_event(
@@ -133,120 +160,64 @@ def common_chat_logic(scope, pricing_info, environment, application_name, metric
},
)
- choice_event_body = {
- 'finish_reason': scope._finish_reason,
- 'index': 0,
- 'message': {
- **({'content': scope._llmresponse} if capture_message_content else {}),
- 'role': scope._response_role
- }
- }
-
- # Emit events
- for role in ['user', 'system', 'assistant', 'tool']:
- if formatted_messages.get(role, {}).get('content', ''):
- event = otel_event(
- name=getattr(SemanticConvention, f'GEN_AI_{role.upper()}_MESSAGE'),
- attributes={
- SemanticConvention.GEN_AI_SYSTEM: SemanticConvention.GEN_AI_SYSTEM_AWS_BEDROCK
- },
- body = {
- # pylint: disable=line-too-long
- **({'content': formatted_messages.get(role, {}).get('content', '')} if capture_message_content else {}),
- 'role': formatted_messages.get(role, {}).get('role', []),
- **({
- 'tool_calls': {
- 'function': {
- # pylint: disable=line-too-long
- 'name': (scope._tool_calls[0].get('function', {}).get('name', '') if scope._tool_calls else ''),
- 'arguments': (scope._tool_calls[0].get('function', {}).get('arguments', '') if scope._tool_calls else '')
- },
- 'id': (scope._tool_calls[0].get('id', '') if scope._tool_calls else ''),
- 'type': 'function'
- }
- } if role == 'assistant' else {}),
- **({
- 'id': (scope._tool_calls[0].get('id', '') if scope._tool_calls else '')
- } if role == 'tool' else {})
- }
- )
- event_provider.emit(event)
-
- choice_event = otel_event(
- name=SemanticConvention.GEN_AI_CHOICE,
- attributes={
- SemanticConvention.GEN_AI_SYSTEM: SemanticConvention.GEN_AI_SYSTEM_AWS_BEDROCK
- },
- body=choice_event_body
- )
- event_provider.emit(choice_event)
-
scope._span.set_status(Status(StatusCode.OK))
+ # Record metrics
if not disable_metrics:
- metrics_attributes = create_metrics_attributes(
- service_name=application_name,
- deployment_environment=environment,
- operation=SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT,
- system=SemanticConvention.GEN_AI_SYSTEM_AWS_BEDROCK,
- request_model=request_model,
- server_address=scope._server_address,
- server_port=scope._server_port,
- response_model=scope._response_model,
- )
-
- metrics['genai_client_usage_tokens'].record(scope._input_tokens + scope._output_tokens, metrics_attributes)
- metrics['genai_client_operation_duration'].record(scope._end_time - scope._start_time, metrics_attributes)
- metrics['genai_server_tbt'].record(scope._tbt, metrics_attributes)
- metrics['genai_server_ttft'].record(scope._ttft, metrics_attributes)
- metrics['genai_requests'].add(1, metrics_attributes)
- metrics['genai_completion_tokens'].add(scope._output_tokens, metrics_attributes)
- metrics['genai_prompt_tokens'].add(scope._input_tokens, metrics_attributes)
- metrics['genai_cost'].record(cost, metrics_attributes)
-
-def process_streaming_chat_response(self, pricing_info, environment, application_name, metrics,
- event_provider, capture_message_content=False, disable_metrics=False, version='', llm_config=''):
+ record_completion_metrics(metrics, SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_AWS_BEDROCK,
+ scope._server_address, scope._server_port, request_model, scope._response_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="", llm_config=None):
"""
- Process chat request and generate Telemetry
+ Process streaming chat response and generate telemetry.
"""
- if self._tool_id != '':
- self._tool_calls = {
- 'id': self._tool_id,
- 'name': self._tool_name,
- 'input': self._tool_arguments
- }
- common_chat_logic(self, pricing_info, environment, application_name, metrics,
- event_provider, capture_message_content, disable_metrics, version, llm_config, is_stream=True)
+ try:
+ if llm_config is None:
+ llm_config = {}
-def process_chat_response(response, request_model, pricing_info, server_port, server_address, environment,
- application_name, metrics, event_provider, start_time, span, capture_message_content=False,
- disable_metrics=False, version='1.0.0', llm_config='', **kwargs):
+ common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+ capture_message_content, disable_metrics, version, llm_config, is_stream=True)
+ except Exception as e:
+ handle_exception(scope._span, e)
+ raise
+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", llm_config=None, **kwargs):
"""
- Process chat request and generate Telemetry
+ Process non-streaming chat response and generate telemetry.
"""
- self = type('GenericScope', (), {})()
- response_dict = response_as_dict(response)
-
- # pylint: disable = no-member
- self._start_time = start_time
- self._end_time = time.time()
- self._span = span
- self._llmresponse = response_dict.get('output').get('message').get('content')[0].get('text')
- self._response_role = 'assistant'
- self._input_tokens = response_dict.get('usage').get('inputTokens')
- self._output_tokens = response_dict.get('usage').get('outputTokens')
- self._response_model = request_model
- self._finish_reason = response_dict.get('stopReason', '')
- self._response_id = response_dict.get('ResponseMetadata').get('RequestId')
- self._timestamps = []
- self._ttft, self._tbt = self._end_time - self._start_time, 0
- self._server_address, self._server_port = server_address, server_port
- self._kwargs = kwargs
- common_chat_logic(self, pricing_info, environment, application_name, metrics,
- event_provider, capture_message_content, disable_metrics, version, llm_config, is_stream=False)
-
- return response
+ try:
+ if llm_config is None:
+ llm_config = {}
+
+ scope = type("GenericScope", (), {})()
+ response_dict = response_as_dict(response)
+
+ scope._start_time = start_time
+ scope._end_time = time.time()
+ scope._span = span
+ scope._llmresponse = response_dict.get("output", {}).get("message", {}).get("content", [{}])[0].get("text", "")
+ scope._response_role = response_dict.get("output", {}).get("message", {}).get("role", "assistant")
+ scope._input_tokens = response_dict.get("usage", {}).get("inputTokens", 0)
+ scope._output_tokens = response_dict.get("usage", {}).get("outputTokens", 0)
+ scope._response_model = request_model
+ scope._finish_reason = response_dict.get("stopReason", "")
+ scope._response_id = response_dict.get("RequestId", "")
+ scope._timestamps = []
+ scope._ttft, scope._tbt = scope._end_time - scope._start_time, 0
+ scope._server_address, scope._server_port = server_address, server_port
+ scope._kwargs = kwargs
+
+ common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+ capture_message_content, disable_metrics, version, llm_config, is_stream=False)
+
+ return response
+ except Exception as e:
+ handle_exception(span, e)
+ raise
diff --git a/sdk/python/src/openlit/instrumentation/litellm/async_litellm.py b/sdk/python/src/openlit/instrumentation/litellm/async_litellm.py
index db21cdf5..9cbfd63a 100644
--- a/sdk/python/src/openlit/instrumentation/litellm/async_litellm.py
+++ b/sdk/python/src/openlit/instrumentation/litellm/async_litellm.py
@@ -68,9 +68,9 @@ def acompletion(version, environment, application_name, tracer, pricing_info,
def __aiter__(self):
return self
- def __getattr__(self, name):
+ async def __getattr__(self, name):
"""Delegate attribute access to the wrapped object."""
- return getattr(self.__wrapped__, name)
+ return getattr(await self.__wrapped__, name)
async def __anext__(self):
try: