openlit 1.33.17__py3-none-any.whl → 1.33.19__py3-none-any.whl

openlit/instrumentation/azure_ai_inference/utils.py

@@ -0,0 +1,225 @@
+"""
+Azure AI Inference 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
+)
+from openlit.semcov import SemanticConvetion
+
+def process_chunk(self, chunk):
+    """
+    Process a chunk of response data and update state.
+    """
+
+    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)
+
+    chunked = response_as_dict(chunk)
+
+    # Collect message IDs and aggregated response from events
+    if (len(chunked.get('choices')) > 0 and ('delta' in chunked.get('choices')[0] and
+        'content' in chunked.get('choices')[0].get('delta'))):
+
+        if content := chunked.get('choices')[0].get('delta').get('content'):
+            self._llmresponse += content
+
+    if chunked.get('choices')[0].get('finish_reason') is not None:
+        self._finish_reason = chunked.get('choices')[0].get('finish_reason')
+
+    if chunked.get('usage') is not None:
+        self._input_tokens = chunked.get('usage').get('prompt_tokens')
+        self._response_id = chunked.get('id')
+        self._response_model = chunked.get('model')
+        self._output_tokens = chunked.get('usage').get('completion_tokens')
+
+def common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+                        event_provider, 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)
+
+    formatted_messages = extract_and_format_input(scope._kwargs.get('messages', ''))
+    request_model = scope._kwargs.get('model', 'claude-3-opus-20240229')
+
+    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(SemanticConvetion.GEN_AI_OPERATION, SemanticConvetion.GEN_AI_OPERATION_TYPE_CHAT)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM, SemanticConvetion.GEN_AI_SYSTEM_AZURE_AI_INFERENCE)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL, request_model)
+    scope._span.set_attribute(SemanticConvetion.SERVER_PORT, scope._server_port)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MAX_TOKENS, scope._kwargs.get('max_tokens', -1))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_STOP_SEQUENCES, scope._kwargs.get('stop', []))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TEMPERATURE, scope._kwargs.get('temperature', 1.0))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_K, scope._kwargs.get('top_k', 1.0))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_P, scope._kwargs.get('top_p', 1.0))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_FREQUENCY_PENALTY,
+                                                          scope._kwargs.get('frequency_penalty', 0.0))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_PRESENCE_PENALTY,
+                                                          scope._kwargs.get('presence_penalty', 0.0))
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_FINISH_REASON, [scope._finish_reason])
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_ID, scope._response_id)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_MODEL, scope._response_model)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_USAGE_INPUT_TOKENS, scope._input_tokens)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_USAGE_OUTPUT_TOKENS, scope._output_tokens)
+    scope._span.set_attribute(SemanticConvetion.SERVER_ADDRESS, scope._server_address)
+
+    scope._span.set_attribute(SemanticConvetion.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(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM, is_stream)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_CLIENT_TOKEN_USAGE, scope._input_tokens + scope._output_tokens)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST, cost)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_SERVER_TBT, scope._tbt)
+    scope._span.set_attribute(SemanticConvetion.GEN_AI_SERVER_TTFT, scope._ttft)
+    scope._span.set_attribute(SemanticConvetion.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)
+    if capture_message_content:
+        scope._span.add_event(
+            name=SemanticConvetion.GEN_AI_CONTENT_PROMPT_EVENT,
+            attributes={
+                SemanticConvetion.GEN_AI_CONTENT_PROMPT: prompt,
+            },
+        )
+        scope._span.add_event(
+            name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,
+            attributes={
+                SemanticConvetion.GEN_AI_CONTENT_COMPLETION: scope._llmresponse,
+            },
+        )
+
+    choice_event_body = {
+        'finish_reason': scope._finish_reason,
+        'index': 0,
+        'message': {
+            **({'content': scope._llmresponse} if capture_message_content else {}),
+            'role': 'assistant'
+        }
+    }
+
+    # Emit events
+    for role in ['user', 'system', 'assistant', 'tool']:
+        if formatted_messages.get(role, {}).get('content', ''):
+            event = otel_event(
+                name=getattr(SemanticConvetion, f'GEN_AI_{role.upper()}_MESSAGE'),
+                attributes={
+                    SemanticConvetion.GEN_AI_SYSTEM: SemanticConvetion.GEN_AI_SYSTEM_AZURE_AI_INFERENCE
+                },
+                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=SemanticConvetion.GEN_AI_CHOICE,
+        attributes={
+            SemanticConvetion.GEN_AI_SYSTEM: SemanticConvetion.GEN_AI_SYSTEM_AZURE_AI_INFERENCE
+        },
+        body=choice_event_body
+    )
+    event_provider.emit(choice_event)
+
+    scope._span.set_status(Status(StatusCode.OK))
+
+    if not disable_metrics:
+        metrics_attributes = create_metrics_attributes(
+            service_name=application_name,
+            deployment_environment=environment,
+            operation=SemanticConvetion.GEN_AI_OPERATION_TYPE_CHAT,
+            system=SemanticConvetion.GEN_AI_SYSTEM_AZURE_AI_INFERENCE,
+            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=''):
+    """
+    Process chat request and generate Telemetry
+    """
+
+    common_chat_logic(self, pricing_info, environment, application_name, metrics,
+                        event_provider, 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, event_provider, start_time,
+                          span, capture_message_content=False, disable_metrics=False, version='1.0.0', **kwargs):
+    """
+    Process chat request 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('choices', {})[0].get('message', '').get('content', '')
+    self._input_tokens = response_dict.get('usage').get('prompt_tokens')
+    self._output_tokens = response_dict.get('usage').get('completion_tokens')
+    self._response_model = response_dict.get('model', '')
+    self._finish_reason = response_dict.get('choices', {})[0].get('finish_reason', '')
+    self._response_id = response_dict.get('id', '')
+    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, is_stream=False)
+
+    return response

openlit/instrumentation/bedrock/__init__.py

@@ -22,6 +22,7 @@ class BedrockInstrumentor(BaseInstrumentor):
         application_name = kwargs.get("application_name", "default_application")
         environment = kwargs.get("environment", "default_environment")
         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)
@@ -33,7 +34,7 @@ class BedrockInstrumentor(BaseInstrumentor):
             "botocore.client",  
             "ClientCreator.create_client",  
             converse(version, environment, application_name,
-                     tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+                     tracer, event_provider, pricing_info, capture_message_content, metrics, disable_metrics),
         )
 
     def _uninstrument(self, **kwargs):

openlit/instrumentation/bedrock/bedrock.py

@@ -4,96 +4,33 @@ Module for monitoring Amazon Bedrock API calls.
 
 import logging
 import time
-from botocore.response import StreamingBody
-from botocore.exceptions import ReadTimeoutError, ResponseStreamingError
-from urllib3.exceptions import ProtocolError as URLLib3ProtocolError
-from urllib3.exceptions import ReadTimeoutError as URLLib3ReadTimeoutError
-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,
-    response_as_dict,
-    create_metrics_attributes,
     set_server_address_and_port
 )
+from openlit.instrumentation.bedrock.utils import (
+    process_chat_response,
+)
 from openlit.semcov import SemanticConvetion
 
 # Initialize logger for logging potential issues and operations
 logger = logging.getLogger(__name__)
 
-class CustomStreamWrapper(StreamingBody):
-    """Handle streaming responses with the ability to read multiple times."""
-
-    def __init__(self, stream_source, length):
-        super().__init__(stream_source, length)
-        self._stream_data = None
-        self._read_position = 0
-
-    def read(self, amt=None):
-        if self._stream_data is None:
-            try:
-                self._stream_data = self._raw_stream.read()
-            except URLLib3ReadTimeoutError as error:
-                raise ReadTimeoutError(endpoint_url=error.url, error=error) from error
-            except URLLib3ProtocolError as error:
-                raise ResponseStreamingError(error=error) from error
-
-            self._amount_read += len(self._stream_data)
-            if amt is None or (not self._stream_data and amt > 0):
-                self._verify_content_length()
-
-        if amt is None:
-            data_chunk = self._stream_data[self._read_position:]
-        else:
-            data_start = self._read_position
-            self._read_position += amt
-            data_chunk = self._stream_data[data_start:self._read_position]
-
-        return data_chunk
-
-def converse(version, environment, application_name, tracer,
+def converse(version, environment, application_name, tracer, event_provider,
          pricing_info, capture_message_content, metrics, disable_metrics):
     """
-    Generates a telemetry wrapper for messages to collect metrics.
-
-    Args:
-        gen_ai_endpoint: Endpoint identifier for logging and tracing.
-        version: The monitoring package version.
-        environment: Deployment environment (e.g. production, staging).
-        application_name: Name of the application using the Bedrock API.
-        tracer: OpenTelemetry tracer for creating spans.
-        pricing_info: Information for calculating Bedrock usage cost.
-        capture_message_content: Whether to trace the actual content.
-        metrics: Metrics collector.
-        disable_metrics: Flag to toggle metrics collection.
-    Returns:
-        A function that wraps the chat method to add telemetry.
+    Generates a telemetry wrapper for GenAI function call
     """
 
     def wrapper(wrapped, instance, args, kwargs):
         """
-        Wraps an API call to add telemetry.
-
-        Args:
-            wrapped: Original method.
-            instance: Instance of the class.
-            args: Positional arguments of the 'messages' method.
-            kwargs: Keyword arguments of the 'messages' method.
-        Returns:
-            Response from the original method.
+        Wraps the GenAI function call.
         """
 
         def converse_wrapper(original_method, *method_args, **method_kwargs):
-            """
-            Adds instrumentation to the invoke model call.
 
-            Args:
-                original_method: The original invoke model method.
-                *method_args: Positional arguments for the method.
-                **method_kwargs: Keyword arguments for the method.
-            Returns:
-                The modified response with telemetry.
+            """
+            Wraps the GenAI function call.
             """
 
             server_address, server_port = set_server_address_and_port(instance, 'aws.amazon.com', 443)
@@ -104,146 +41,27 @@ def converse(version, environment, application_name, tracer,
             with tracer.start_as_current_span(span_name, kind=SpanKind.CLIENT) as span:
                 start_time = time.time()
                 response = original_method(*method_args, **method_kwargs)
-                end_time = time.time()
-
-                response_dict = response_as_dict(response)
-
-                try:
-                    message_prompt = method_kwargs.get('messages', '')
-                    formatted_messages = []
-                    for message in message_prompt:
-                        role = message['role']
-                        content = message['content']
-
-                        if isinstance(content, list):
-                            content_str = ", ".join(f'text: {item["text"]}' for item in content if "text" in item)
-                            formatted_messages.append(f'{role}: {content_str}')
-                        else:
-                            formatted_messages.append(f'{role}: {content}')
-                    prompt = '\n'.join(formatted_messages)
-
-                    input_tokens = response_dict.get('usage').get('inputTokens')
-                    output_tokens = response_dict.get('usage').get('outputTokens')
-
-                    # Calculate cost of the operation
-                    cost = get_chat_model_cost(request_model, pricing_info,
-                                                input_tokens, output_tokens)
-
-                    llm_response = response_dict.get('output').get('message').get('content')[0].get('text')
-
-                    # Set base span attribues (OTel Semconv)
-                    span.set_attribute(TELEMETRY_SDK_NAME, 'openlit')
-                    span.set_attribute(SemanticConvetion.GEN_AI_OPERATION,
-                                        SemanticConvetion.GEN_AI_OPERATION_TYPE_CHAT)
-                    span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
-                                        SemanticConvetion.GEN_AI_SYSTEM_AWS_BEDROCK)
-                    span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL,
-                                        request_model)
-                    span.set_attribute(SemanticConvetion.SERVER_PORT,
-                                        server_port)
-
-                    inference_config = method_kwargs.get('inferenceConfig', {})
-
-                    # List of attributes and their config keys
-                    attributes = [
-                        (SemanticConvetion.GEN_AI_REQUEST_FREQUENCY_PENALTY, 'frequencyPenalty'),
-                        (SemanticConvetion.GEN_AI_REQUEST_MAX_TOKENS, 'maxTokens'),
-                        (SemanticConvetion.GEN_AI_REQUEST_PRESENCE_PENALTY, 'presencePenalty'),
-                        (SemanticConvetion.GEN_AI_REQUEST_STOP_SEQUENCES, 'stopSequences'),
-                        (SemanticConvetion.GEN_AI_REQUEST_TEMPERATURE, 'temperature'),
-                        (SemanticConvetion.GEN_AI_REQUEST_TOP_P, 'topP'),
-                        (SemanticConvetion.GEN_AI_REQUEST_TOP_K, 'topK'),
-                    ]
-
-                    # 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:
-                            span.set_attribute(attribute, value)
-
-                    span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_ID,
-                                        response_dict.get('ResponseMetadata').get('RequestId'))
-                    span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_MODEL,
-                                        request_model)
-                    span.set_attribute(SemanticConvetion.GEN_AI_USAGE_INPUT_TOKENS,
-                                        input_tokens)
-                    span.set_attribute(SemanticConvetion.GEN_AI_USAGE_OUTPUT_TOKENS,
-                                        output_tokens)
-                    span.set_attribute(SemanticConvetion.SERVER_ADDRESS,
-                                        server_address)
-                    if isinstance(llm_response, str):
-                        span.set_attribute(SemanticConvetion.GEN_AI_OUTPUT_TYPE,
-                                        'text')
-                    else:
-                        span.set_attribute(SemanticConvetion.GEN_AI_OUTPUT_TYPE,
-                                        'json')
-
-                    # Set base span attribues (Extras)
-                    span.set_attribute(DEPLOYMENT_ENVIRONMENT,
-                                        environment)
-                    span.set_attribute(SERVICE_NAME,
-                                        application_name)
-                    span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
-                                        False)
-                    span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
-                                        input_tokens + output_tokens)
-                    span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
-                                        cost)
-                    span.set_attribute(SemanticConvetion.GEN_AI_SERVER_TTFT,
-                                        end_time - start_time)
-                    span.set_attribute(SemanticConvetion.GEN_AI_SDK_VERSION,
-                                        version)
-
-                    if capture_message_content:
-                        span.add_event(
-                            name=SemanticConvetion.GEN_AI_CONTENT_PROMPT_EVENT,
-                            attributes={
-                                SemanticConvetion.GEN_AI_CONTENT_PROMPT: prompt,
-                            },
-                        )
-                        span.add_event(
-                            name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,
-                            attributes={
-                                SemanticConvetion.GEN_AI_CONTENT_COMPLETION: llm_response,
-                            },
-                        )
-
-                    span.set_status(Status(StatusCode.OK))
-
-                    if disable_metrics is False:
-                        attributes = create_metrics_attributes(
-                            service_name=application_name,
-                            deployment_environment=environment,
-                            operation=SemanticConvetion.GEN_AI_OPERATION_TYPE_CHAT,
-                            system=SemanticConvetion.GEN_AI_SYSTEM_AWS_BEDROCK,
-                            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 response
-
-                except Exception as e:
-                    handle_exception(span, e)
-                    logger.error('Error in trace creation: %s', e)
-
-                    # Return original response
-                    return response
+                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
+                )
+
+                return response
 
         # Get the original client instance from the wrapper
         client = wrapped(*args, **kwargs)