sentry-sdk 2.41.0__py2.py3-none-any.whl → 2.42.1__py2.py3-none-any.whl

sentry_sdk/ai/utils.py

@@ -1,14 +1,18 @@
 import json
-
+from collections import deque
 from typing import TYPE_CHECKING
+from sys import getsizeof
 
 if TYPE_CHECKING:
-    from typing import Any, Callable
+    from typing import Any, Callable, Dict, List, Optional, Tuple
+
     from sentry_sdk.tracing import Span
 
 import sentry_sdk
 from sentry_sdk.utils import logger
 
+MAX_GEN_AI_MESSAGE_BYTES = 20_000  # 20KB
+
 
 class GEN_AI_ALLOWED_MESSAGE_ROLES:
     SYSTEM = "system"
@@ -95,3 +99,46 @@ def get_start_span_function():
         current_span is not None and current_span.containing_transaction is not None
     )
     return sentry_sdk.start_span if transaction_exists else sentry_sdk.start_transaction
+
+
+def _find_truncation_index(messages, max_bytes):
+    # type: (List[Dict[str, Any]], int) -> int
+    """
+    Find the index of the first message that would exceed the max bytes limit.
+    Compute the individual message sizes, and return the index of the first message from the back
+    of the list that would exceed the max bytes limit.
+    """
+    running_sum = 0
+    for idx in range(len(messages) - 1, -1, -1):
+        size = len(json.dumps(messages[idx], separators=(",", ":")).encode("utf-8"))
+        running_sum += size
+        if running_sum > max_bytes:
+            return idx + 1
+
+    return 0
+
+
+def truncate_messages_by_size(messages, max_bytes=MAX_GEN_AI_MESSAGE_BYTES):
+    # type: (List[Dict[str, Any]], int) -> Tuple[List[Dict[str, Any]], int]
+    serialized_json = json.dumps(messages, separators=(",", ":"))
+    current_size = len(serialized_json.encode("utf-8"))
+
+    if current_size <= max_bytes:
+        return messages, 0
+
+    truncation_index = _find_truncation_index(messages, max_bytes)
+    return messages[truncation_index:], truncation_index
+
+
+def truncate_and_annotate_messages(
+    messages, span, scope, max_bytes=MAX_GEN_AI_MESSAGE_BYTES
+):
+    # type: (Optional[List[Dict[str, Any]]], Any, Any, int) -> Optional[List[Dict[str, Any]]]
+    if not messages:
+        return None
+
+    truncated_messages, removed_count = truncate_messages_by_size(messages, max_bytes)
+    if removed_count > 0:
+        scope._gen_ai_original_message_count[span.span_id] = len(messages)
+
+    return truncated_messages

sentry_sdk/client.py

@@ -598,14 +598,31 @@ class _Client(BaseClient):
             if event_scrubber:
                 event_scrubber.scrub_event(event)
 
+        if scope is not None and scope._gen_ai_original_message_count:
+            spans = event.get("spans", [])  # type: List[Dict[str, Any]] | AnnotatedValue
+            if isinstance(spans, list):
+                for span in spans:
+                    span_id = span.get("span_id", None)
+                    span_data = span.get("data", {})
+                    if (
+                        span_id
+                        and span_id in scope._gen_ai_original_message_count
+                        and SPANDATA.GEN_AI_REQUEST_MESSAGES in span_data
+                    ):
+                        span_data[SPANDATA.GEN_AI_REQUEST_MESSAGES] = AnnotatedValue(
+                            span_data[SPANDATA.GEN_AI_REQUEST_MESSAGES],
+                            {"len": scope._gen_ai_original_message_count[span_id]},
+                        )
         if previous_total_spans is not None:
             event["spans"] = AnnotatedValue(
                 event.get("spans", []), {"len": previous_total_spans}
             )
         if previous_total_breadcrumbs is not None:
             event["breadcrumbs"] = AnnotatedValue(
-                event.get("breadcrumbs", []), {"len": previous_total_breadcrumbs}
+                event.get("breadcrumbs", {"values": []}),
+                {"len": previous_total_breadcrumbs},
             )
+
         # Postprocess the event here so that annotated types do
         # generally not surface in before_send
         if event is not None:

sentry_sdk/consts.py

@@ -909,6 +909,8 @@ class ClientConstructor:
         error_sampler=None,  # type: Optional[Callable[[Event, Hint], Union[float, bool]]]
         enable_db_query_source=True,  # type: bool
         db_query_source_threshold_ms=100,  # type: int
+        enable_http_request_source=False,  # type: bool
+        http_request_source_threshold_ms=100,  # type: int
         spotlight=None,  # type: Optional[Union[bool, str]]
         cert_file=None,  # type: Optional[str]
         key_file=None,  # type: Optional[str]
@@ -1264,6 +1266,13 @@ class ClientConstructor:
 
             The query location will be added to the query for queries slower than the specified threshold.
 
+        :param enable_http_request_source: When enabled, the source location will be added to outgoing HTTP requests.
+
+        :param http_request_source_threshold_ms: The threshold in milliseconds for adding the source location to an
+            outgoing HTTP request.
+
+            The request location will be added to the request for requests slower than the specified threshold.
+
         :param custom_repr: A custom `repr <https://docs.python.org/3/library/functions.html#repr>`_ function to run
             while serializing an object.
 
@@ -1339,4 +1348,4 @@ DEFAULT_OPTIONS = _get_default_options()
 del _get_default_options
 
 
-VERSION = "2.41.0"
+VERSION = "2.42.1"

sentry_sdk/integrations/__init__.py

@@ -140,6 +140,7 @@ _MIN_VERSIONS = {
     "flask": (1, 1, 4),
     "gql": (3, 4, 1),
     "graphene": (3, 3),
+    "google_genai": (1, 29, 0),  # google-genai
     "grpc": (1, 32, 0),  # grpcio
     "httpx": (0, 16, 0),
     "huggingface_hub": (0, 24, 7),

sentry_sdk/integrations/aiohttp.py

@@ -22,7 +22,7 @@ from sentry_sdk.tracing import (
     SOURCE_FOR_STYLE,
     TransactionSource,
 )
-from sentry_sdk.tracing_utils import should_propagate_trace
+from sentry_sdk.tracing_utils import should_propagate_trace, add_http_request_source
 from sentry_sdk.utils import (
     capture_internal_exceptions,
     ensure_integration_enabled,
@@ -279,6 +279,9 @@ def create_trace_config():
         span.set_data("reason", params.response.reason)
         span.finish()
 
+        with capture_internal_exceptions():
+            add_http_request_source(span)
+
     trace_config = TraceConfig()
 
     trace_config.on_request_start.append(on_request_start)

sentry_sdk/integrations/anthropic.py

@@ -6,6 +6,7 @@ from sentry_sdk.ai.monitoring import record_token_usage
 from sentry_sdk.ai.utils import (
     set_data_normalized,
     normalize_message_roles,
+    truncate_and_annotate_messages,
     get_start_span_function,
 )
 from sentry_sdk.consts import OP, SPANDATA, SPANSTATUS
@@ -145,12 +146,14 @@ def _set_input_data(span, kwargs, integration):
                 normalized_messages.append(message)
 
         role_normalized_messages = normalize_message_roles(normalized_messages)
-        set_data_normalized(
-            span,
-            SPANDATA.GEN_AI_REQUEST_MESSAGES,
-            role_normalized_messages,
-            unpack=False,
+        scope = sentry_sdk.get_current_scope()
+        messages_data = truncate_and_annotate_messages(
+            role_normalized_messages, span, scope
         )
+        if messages_data is not None:
+            set_data_normalized(
+                span, SPANDATA.GEN_AI_REQUEST_MESSAGES, messages_data, unpack=False
+            )
 
     set_data_normalized(
         span, SPANDATA.GEN_AI_RESPONSE_STREAMING, kwargs.get("stream", False)

sentry_sdk/integrations/aws_lambda.py

@@ -138,6 +138,8 @@ def _wrap_handler(handler):
                     timeout_thread = TimeoutThread(
                         waiting_time,
                         configured_time / MILLIS_TO_SECONDS,
+                        isolation_scope=scope,
+                        current_scope=sentry_sdk.get_current_scope(),
                     )
 
                     # Starting the thread to raise timeout warning exception

sentry_sdk/integrations/gcp.py

@@ -75,7 +75,12 @@ def _wrap_func(func):
                 ):
                     waiting_time = configured_time - TIMEOUT_WARNING_BUFFER
 
-                    timeout_thread = TimeoutThread(waiting_time, configured_time)
+                    timeout_thread = TimeoutThread(
+                        waiting_time,
+                        configured_time,
+                        isolation_scope=scope,
+                        current_scope=sentry_sdk.get_current_scope(),
+                    )
 
                     # Starting the thread to raise timeout warning exception
                     timeout_thread.start()

sentry_sdk/integrations/google_genai/__init__.py

@@ -0,0 +1,298 @@
+from functools import wraps
+from typing import (
+    Any,
+    AsyncIterator,
+    Callable,
+    Iterator,
+    List,
+)
+
+import sentry_sdk
+from sentry_sdk.ai.utils import get_start_span_function
+from sentry_sdk.integrations import DidNotEnable, Integration
+from sentry_sdk.consts import OP, SPANDATA
+from sentry_sdk.tracing import SPANSTATUS
+
+
+try:
+    from google.genai.models import Models, AsyncModels
+except ImportError:
+    raise DidNotEnable("google-genai not installed")
+
+
+from .consts import IDENTIFIER, ORIGIN, GEN_AI_SYSTEM
+from .utils import (
+    set_span_data_for_request,
+    set_span_data_for_response,
+    _capture_exception,
+    prepare_generate_content_args,
+)
+from .streaming import (
+    set_span_data_for_streaming_response,
+    accumulate_streaming_response,
+)
+
+
+class GoogleGenAIIntegration(Integration):
+    identifier = IDENTIFIER
+    origin = ORIGIN
+
+    def __init__(self, include_prompts=True):
+        # type: (GoogleGenAIIntegration, bool) -> None
+        self.include_prompts = include_prompts
+
+    @staticmethod
+    def setup_once():
+        # type: () -> None
+        # Patch sync methods
+        Models.generate_content = _wrap_generate_content(Models.generate_content)
+        Models.generate_content_stream = _wrap_generate_content_stream(
+            Models.generate_content_stream
+        )
+
+        # Patch async methods
+        AsyncModels.generate_content = _wrap_async_generate_content(
+            AsyncModels.generate_content
+        )
+        AsyncModels.generate_content_stream = _wrap_async_generate_content_stream(
+            AsyncModels.generate_content_stream
+        )
+
+
+def _wrap_generate_content_stream(f):
+    # type: (Callable[..., Any]) -> Callable[..., Any]
+    @wraps(f)
+    def new_generate_content_stream(self, *args, **kwargs):
+        # type: (Any, Any, Any) -> Any
+        integration = sentry_sdk.get_client().get_integration(GoogleGenAIIntegration)
+        if integration is None:
+            return f(self, *args, **kwargs)
+
+        _model, contents, model_name = prepare_generate_content_args(args, kwargs)
+
+        span = get_start_span_function()(
+            op=OP.GEN_AI_INVOKE_AGENT,
+            name="invoke_agent",
+            origin=ORIGIN,
+        )
+        span.__enter__()
+        span.set_data(SPANDATA.GEN_AI_AGENT_NAME, model_name)
+        span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
+        set_span_data_for_request(span, integration, model_name, contents, kwargs)
+        span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)
+
+        chat_span = sentry_sdk.start_span(
+            op=OP.GEN_AI_CHAT,
+            name=f"chat {model_name}",
+            origin=ORIGIN,
+        )
+        chat_span.__enter__()
+        chat_span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "chat")
+        chat_span.set_data(SPANDATA.GEN_AI_SYSTEM, GEN_AI_SYSTEM)
+        chat_span.set_data(SPANDATA.GEN_AI_REQUEST_MODEL, model_name)
+        set_span_data_for_request(chat_span, integration, model_name, contents, kwargs)
+        chat_span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)
+
+        try:
+            stream = f(self, *args, **kwargs)
+
+            # Create wrapper iterator to accumulate responses
+            def new_iterator():
+                # type: () -> Iterator[Any]
+                chunks = []  # type: List[Any]
+                try:
+                    for chunk in stream:
+                        chunks.append(chunk)
+                        yield chunk
+                except Exception as exc:
+                    _capture_exception(exc)
+                    chat_span.set_status(SPANSTATUS.ERROR)
+                    raise
+                finally:
+                    # Accumulate all chunks and set final response data on spans
+                    if chunks:
+                        accumulated_response = accumulate_streaming_response(chunks)
+                        set_span_data_for_streaming_response(
+                            chat_span, integration, accumulated_response
+                        )
+                        set_span_data_for_streaming_response(
+                            span, integration, accumulated_response
+                        )
+                    chat_span.__exit__(None, None, None)
+                    span.__exit__(None, None, None)
+
+            return new_iterator()
+
+        except Exception as exc:
+            _capture_exception(exc)
+            chat_span.__exit__(None, None, None)
+            span.__exit__(None, None, None)
+            raise
+
+    return new_generate_content_stream
+
+
+def _wrap_async_generate_content_stream(f):
+    # type: (Callable[..., Any]) -> Callable[..., Any]
+    @wraps(f)
+    async def new_async_generate_content_stream(self, *args, **kwargs):
+        # type: (Any, Any, Any) -> Any
+        integration = sentry_sdk.get_client().get_integration(GoogleGenAIIntegration)
+        if integration is None:
+            return await f(self, *args, **kwargs)
+
+        _model, contents, model_name = prepare_generate_content_args(args, kwargs)
+
+        span = get_start_span_function()(
+            op=OP.GEN_AI_INVOKE_AGENT,
+            name="invoke_agent",
+            origin=ORIGIN,
+        )
+        span.__enter__()
+        span.set_data(SPANDATA.GEN_AI_AGENT_NAME, model_name)
+        span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
+        set_span_data_for_request(span, integration, model_name, contents, kwargs)
+        span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)
+
+        chat_span = sentry_sdk.start_span(
+            op=OP.GEN_AI_CHAT,
+            name=f"chat {model_name}",
+            origin=ORIGIN,
+        )
+        chat_span.__enter__()
+        chat_span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "chat")
+        chat_span.set_data(SPANDATA.GEN_AI_SYSTEM, GEN_AI_SYSTEM)
+        chat_span.set_data(SPANDATA.GEN_AI_REQUEST_MODEL, model_name)
+        set_span_data_for_request(chat_span, integration, model_name, contents, kwargs)
+        chat_span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)
+
+        try:
+            stream = await f(self, *args, **kwargs)
+
+            # Create wrapper async iterator to accumulate responses
+            async def new_async_iterator():
+                # type: () -> AsyncIterator[Any]
+                chunks = []  # type: List[Any]
+                try:
+                    async for chunk in stream:
+                        chunks.append(chunk)
+                        yield chunk
+                except Exception as exc:
+                    _capture_exception(exc)
+                    chat_span.set_status(SPANSTATUS.ERROR)
+                    raise
+                finally:
+                    # Accumulate all chunks and set final response data on spans
+                    if chunks:
+                        accumulated_response = accumulate_streaming_response(chunks)
+                        set_span_data_for_streaming_response(
+                            chat_span, integration, accumulated_response
+                        )
+                        set_span_data_for_streaming_response(
+                            span, integration, accumulated_response
+                        )
+                    chat_span.__exit__(None, None, None)
+                    span.__exit__(None, None, None)
+
+            return new_async_iterator()
+
+        except Exception as exc:
+            _capture_exception(exc)
+            chat_span.__exit__(None, None, None)
+            span.__exit__(None, None, None)
+            raise
+
+    return new_async_generate_content_stream
+
+
+def _wrap_generate_content(f):
+    # type: (Callable[..., Any]) -> Callable[..., Any]
+    @wraps(f)
+    def new_generate_content(self, *args, **kwargs):
+        # type: (Any, Any, Any) -> Any
+        integration = sentry_sdk.get_client().get_integration(GoogleGenAIIntegration)
+        if integration is None:
+            return f(self, *args, **kwargs)
+
+        model, contents, model_name = prepare_generate_content_args(args, kwargs)
+
+        with get_start_span_function()(
+            op=OP.GEN_AI_INVOKE_AGENT,
+            name="invoke_agent",
+            origin=ORIGIN,
+        ) as span:
+            span.set_data(SPANDATA.GEN_AI_AGENT_NAME, model_name)
+            span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
+            set_span_data_for_request(span, integration, model_name, contents, kwargs)
+
+            with sentry_sdk.start_span(
+                op=OP.GEN_AI_CHAT,
+                name=f"chat {model_name}",
+                origin=ORIGIN,
+            ) as chat_span:
+                chat_span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "chat")
+                chat_span.set_data(SPANDATA.GEN_AI_SYSTEM, GEN_AI_SYSTEM)
+                chat_span.set_data(SPANDATA.GEN_AI_REQUEST_MODEL, model_name)
+                set_span_data_for_request(
+                    chat_span, integration, model_name, contents, kwargs
+                )
+
+                try:
+                    response = f(self, *args, **kwargs)
+                except Exception as exc:
+                    _capture_exception(exc)
+                    chat_span.set_status(SPANSTATUS.ERROR)
+                    raise
+
+                set_span_data_for_response(chat_span, integration, response)
+                set_span_data_for_response(span, integration, response)
+
+                return response
+
+    return new_generate_content
+
+
+def _wrap_async_generate_content(f):
+    # type: (Callable[..., Any]) -> Callable[..., Any]
+    @wraps(f)
+    async def new_async_generate_content(self, *args, **kwargs):
+        # type: (Any, Any, Any) -> Any
+        integration = sentry_sdk.get_client().get_integration(GoogleGenAIIntegration)
+        if integration is None:
+            return await f(self, *args, **kwargs)
+
+        model, contents, model_name = prepare_generate_content_args(args, kwargs)
+
+        with get_start_span_function()(
+            op=OP.GEN_AI_INVOKE_AGENT,
+            name="invoke_agent",
+            origin=ORIGIN,
+        ) as span:
+            span.set_data(SPANDATA.GEN_AI_AGENT_NAME, model_name)
+            span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
+            set_span_data_for_request(span, integration, model_name, contents, kwargs)
+
+            with sentry_sdk.start_span(
+                op=OP.GEN_AI_CHAT,
+                name=f"chat {model_name}",
+                origin=ORIGIN,
+            ) as chat_span:
+                chat_span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "chat")
+                chat_span.set_data(SPANDATA.GEN_AI_SYSTEM, GEN_AI_SYSTEM)
+                chat_span.set_data(SPANDATA.GEN_AI_REQUEST_MODEL, model_name)
+                set_span_data_for_request(
+                    chat_span, integration, model_name, contents, kwargs
+                )
+                try:
+                    response = await f(self, *args, **kwargs)
+                except Exception as exc:
+                    _capture_exception(exc)
+                    chat_span.set_status(SPANSTATUS.ERROR)
+                    raise
+
+                set_span_data_for_response(chat_span, integration, response)
+                set_span_data_for_response(span, integration, response)
+
+                return response
+
+    return new_async_generate_content

sentry_sdk/integrations/google_genai/consts.py

@@ -0,0 +1,16 @@
+GEN_AI_SYSTEM = "gcp.gemini"
+
+# Mapping of tool attributes to their descriptions
+# These are all tools that are available in the Google GenAI API
+TOOL_ATTRIBUTES_MAP = {
+    "google_search_retrieval": "Google Search retrieval tool",
+    "google_search": "Google Search tool",
+    "retrieval": "Retrieval tool",
+    "enterprise_web_search": "Enterprise web search tool",
+    "google_maps": "Google Maps tool",
+    "code_execution": "Code execution tool",
+    "computer_use": "Computer use tool",
+}
+
+IDENTIFIER = "google_genai"
+ORIGIN = f"auto.ai.{IDENTIFIER}"

sentry_sdk/integrations/google_genai/streaming.py

@@ -0,0 +1,155 @@
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    List,
+    TypedDict,
+    Optional,
+)
+
+from sentry_sdk.ai.utils import set_data_normalized
+from sentry_sdk.consts import SPANDATA
+from sentry_sdk.scope import should_send_default_pii
+from sentry_sdk.utils import (
+    safe_serialize,
+)
+from .utils import (
+    extract_tool_calls,
+    extract_finish_reasons,
+    extract_contents_text,
+    extract_usage_data,
+    UsageData,
+)
+
+if TYPE_CHECKING:
+    from sentry_sdk.tracing import Span
+    from google.genai.types import GenerateContentResponse
+
+
+class AccumulatedResponse(TypedDict):
+    id: Optional[str]
+    model: Optional[str]
+    text: str
+    finish_reasons: List[str]
+    tool_calls: List[dict[str, Any]]
+    usage_metadata: UsageData
+
+
+def accumulate_streaming_response(chunks):
+    # type: (List[GenerateContentResponse]) -> AccumulatedResponse
+    """Accumulate streaming chunks into a single response-like object."""
+    accumulated_text = []
+    finish_reasons = []
+    tool_calls = []
+    total_input_tokens = 0
+    total_output_tokens = 0
+    total_tokens = 0
+    total_cached_tokens = 0
+    total_reasoning_tokens = 0
+    response_id = None
+    model = None
+
+    for chunk in chunks:
+        # Extract text and tool calls
+        if getattr(chunk, "candidates", None):
+            for candidate in getattr(chunk, "candidates", []):
+                if hasattr(candidate, "content") and getattr(
+                    candidate.content, "parts", []
+                ):
+                    extracted_text = extract_contents_text(candidate.content)
+                    if extracted_text:
+                        accumulated_text.append(extracted_text)
+
+        extracted_finish_reasons = extract_finish_reasons(chunk)
+        if extracted_finish_reasons:
+            finish_reasons.extend(extracted_finish_reasons)
+
+        extracted_tool_calls = extract_tool_calls(chunk)
+        if extracted_tool_calls:
+            tool_calls.extend(extracted_tool_calls)
+
+        # Accumulate token usage
+        extracted_usage_data = extract_usage_data(chunk)
+        total_input_tokens += extracted_usage_data["input_tokens"]
+        total_output_tokens += extracted_usage_data["output_tokens"]
+        total_cached_tokens += extracted_usage_data["input_tokens_cached"]
+        total_reasoning_tokens += extracted_usage_data["output_tokens_reasoning"]
+        total_tokens += extracted_usage_data["total_tokens"]
+
+    accumulated_response = AccumulatedResponse(
+        text="".join(accumulated_text),
+        finish_reasons=finish_reasons,
+        tool_calls=tool_calls,
+        usage_metadata=UsageData(
+            input_tokens=total_input_tokens,
+            output_tokens=total_output_tokens,
+            input_tokens_cached=total_cached_tokens,
+            output_tokens_reasoning=total_reasoning_tokens,
+            total_tokens=total_tokens,
+        ),
+        id=response_id,
+        model=model,
+    )
+
+    return accumulated_response
+
+
+def set_span_data_for_streaming_response(span, integration, accumulated_response):
+    # type: (Span, Any, AccumulatedResponse) -> None
+    """Set span data for accumulated streaming response."""
+    if (
+        should_send_default_pii()
+        and integration.include_prompts
+        and accumulated_response.get("text")
+    ):
+        span.set_data(
+            SPANDATA.GEN_AI_RESPONSE_TEXT,
+            safe_serialize([accumulated_response["text"]]),
+        )
+
+    if accumulated_response.get("finish_reasons"):
+        set_data_normalized(
+            span,
+            SPANDATA.GEN_AI_RESPONSE_FINISH_REASONS,
+            accumulated_response["finish_reasons"],
+        )
+
+    if accumulated_response.get("tool_calls"):
+        span.set_data(
+            SPANDATA.GEN_AI_RESPONSE_TOOL_CALLS,
+            safe_serialize(accumulated_response["tool_calls"]),
+        )
+
+    if accumulated_response.get("id"):
+        span.set_data(SPANDATA.GEN_AI_RESPONSE_ID, accumulated_response["id"])
+    if accumulated_response.get("model"):
+        span.set_data(SPANDATA.GEN_AI_RESPONSE_MODEL, accumulated_response["model"])
+
+    if accumulated_response["usage_metadata"]["input_tokens"]:
+        span.set_data(
+            SPANDATA.GEN_AI_USAGE_INPUT_TOKENS,
+            accumulated_response["usage_metadata"]["input_tokens"],
+        )
+
+    if accumulated_response["usage_metadata"]["input_tokens_cached"]:
+        span.set_data(
+            SPANDATA.GEN_AI_USAGE_INPUT_TOKENS_CACHED,
+            accumulated_response["usage_metadata"]["input_tokens_cached"],
+        )
+
+    if accumulated_response["usage_metadata"]["output_tokens"]:
+        span.set_data(
+            SPANDATA.GEN_AI_USAGE_OUTPUT_TOKENS,
+            accumulated_response["usage_metadata"]["output_tokens"],
+        )
+
+    if accumulated_response["usage_metadata"]["output_tokens_reasoning"]:
+        span.set_data(
+            SPANDATA.GEN_AI_USAGE_OUTPUT_TOKENS_REASONING,
+            accumulated_response["usage_metadata"]["output_tokens_reasoning"],
+        )
+
+    if accumulated_response["usage_metadata"]["total_tokens"]:
+        span.set_data(
+            SPANDATA.GEN_AI_USAGE_TOTAL_TOKENS,
+            accumulated_response["usage_metadata"]["total_tokens"],
+        )