Skip to content

Python: record gen_ai.client.operation.duration for failed chat and embedding calls #8347

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Evan Mattson (moonbox3) merged 4 commits into from
Sep 15, 2026
+351 −41
Merged

Python: record gen_ai.client.operation.duration for failed chat and embedding calls #8347

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7b1f366
Python: record gen_ai.client.operation.duration for failed chat and e…
rksharma-owg Sep 13, 2026
1ba927b
Python: record duration metric when stream finalizer raises
rksharma-owg Sep 13, 2026
465198a
fix(python): re-raise finalizer and result-hook exceptions in _finali…
rksharma-owg Sep 14, 2026
3cc91b8
Merge main to include Hyperlight CI cleanup fix
rksharma-owg Sep 15, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
144 changes: 103 additions & 41 deletions python/packages/core/agent_framework/observability.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2064,59 +2064,85 @@ def _record_duration() -> None:
)
except Exception as exception:
capture_exception(span=span, exception=exception, timestamp=time_ns())
_capture_operation_error(
attributes=attributes,
exception=exception,
operation_duration_histogram=getattr(self, "duration_histogram", None),
duration=perf_counter() - start_time,
)
_close_span()
raise

async def _finalize_stream() -> None:
from ._types import ChatResponse

try:
if result_stream._stream_error is not None: # pyright: ignore[reportPrivateUsage]
# Stream errored; skip get_final_response() to avoid firing
# result hooks such as after_run context providers on error
# paths. Capture the error on the span before returning.
capture_exception(
span=span,
exception=result_stream._stream_error, # type: ignore
timestamp=time_ns(),
)
return
response: ChatResponse[Any] = await result_stream.get_final_response()
duration = duration_state.get("duration")
response_attributes = _get_response_attributes(attributes, response)
self._backfill_request_model(span, response_attributes)
_capture_response(
if result_stream._stream_error is not None: # pyright: ignore[reportPrivateUsage]
# Stream errored; skip get_final_response() to avoid firing
# result hooks such as after_run context providers on error
# paths. Capture the error on the span before returning.
capture_exception(
span=span,
attributes=response_attributes,
token_usage_histogram=getattr(self, "token_usage_histogram", None),
exception=result_stream._stream_error, # type: ignore
timestamp=time_ns(),
)
_capture_operation_error(
attributes=attributes,
exception=result_stream._stream_error, # type: ignore
operation_duration_histogram=getattr(self, "duration_histogram", None),
duration=duration,
duration=duration_state.get("duration", perf_counter() - start_time),
)
_mark_inner_response_telemetry_captured(response)
if (
OBSERVABILITY_SETTINGS.SENSITIVE_DATA_ENABLED
and isinstance(response, ChatResponse)
and response.messages
and span.is_recording()
):
finish_reason = _get_response_finish_reason(response)
# Activate the span: this cleanup hook runs after the final pull has
# exited its _activate_span context, so it wouldn't otherwise be current.
with _activate_span(span):
_capture_message_events_v1_36(
provider_name=provider_name,
_close_span()
return

try:
try:
response: ChatResponse[Any] = await result_stream.get_final_response()
except Exception as exception:
capture_exception(span=span, exception=exception, timestamp=time_ns())
_capture_operation_error(
attributes=attributes,
exception=exception,
operation_duration_histogram=getattr(self, "duration_histogram", None),
duration=duration_state.get("duration", perf_counter() - start_time),
)
Comment thread
rksharma-owg marked this conversation as resolved.
raise

try:
duration = duration_state.get("duration")
response_attributes = _get_response_attributes(attributes, response)
self._backfill_request_model(span, response_attributes)
_capture_response(
span=span,
attributes=response_attributes,
token_usage_histogram=getattr(self, "token_usage_histogram", None),
operation_duration_histogram=getattr(self, "duration_histogram", None),
duration=duration,
)
_mark_inner_response_telemetry_captured(response)
if (
OBSERVABILITY_SETTINGS.SENSITIVE_DATA_ENABLED
and isinstance(response, ChatResponse)
and response.messages
and span.is_recording()
):
finish_reason = _get_response_finish_reason(response)
# Activate the span: this cleanup hook runs after the final pull has
# exited its _activate_span context, so it wouldn't otherwise be current.
with _activate_span(span):
_capture_message_events_v1_36(
provider_name=provider_name,
messages=response.messages,
finish_reason=finish_reason,
output=True,
)
_capture_message_span_attributes_latest_experimental(
span=span,
messages=response.messages,
finish_reason=finish_reason,
output=True,
)
_capture_message_span_attributes_latest_experimental(
span=span,
messages=response.messages,
finish_reason=finish_reason,
output=True,
)
except Exception as exception:
capture_exception(span=span, exception=exception, timestamp=time_ns())
except Exception as telemetry_exception:
logger.debug("Failed to capture telemetry for stream: %s", telemetry_exception)
finally:
_close_span()

Expand Down Expand Up @@ -2171,6 +2197,12 @@ async def _get_response() -> ChatResponse:
)
except Exception as exception:
capture_exception(span=span, exception=exception, timestamp=time_ns())
_capture_operation_error(
attributes=attributes,
exception=exception,
operation_duration_histogram=getattr(self, "duration_histogram", None),
duration=perf_counter() - start_time_stamp,
)
raise
duration = perf_counter() - start_time_stamp
response_attributes = _get_response_attributes(attributes, response)
Expand Down Expand Up @@ -2256,6 +2288,12 @@ async def get_embeddings(
)
except Exception as exception:
capture_exception(span=span, exception=exception, timestamp=time_ns())
_capture_operation_error(
attributes=attributes,
exception=exception,
operation_duration_histogram=getattr(self, "duration_histogram", None),
duration=perf_counter() - start_time_stamp,
)
raise
duration = perf_counter() - start_time_stamp
response_attributes: dict[str, Any] = {**attributes}
Expand Down Expand Up @@ -3517,6 +3555,30 @@ def _get_response_attributes(
)


def _filter_metric_attributes(attributes: dict[str, Any]) -> dict[str, Any]:
"""Filter attributes to only the GenAI metric attribute set."""
return {key: attributes[key] for key in GEN_AI_METRIC_ATTRIBUTES if key in attributes}


def _capture_operation_error(
attributes: dict[str, Any],
exception: BaseException,
operation_duration_histogram: metrics.Histogram | None = None,
duration: float | None = None,
) -> None:
"""Record the operation duration metric for a call that failed.

The GenAI semantic conventions define ``gen_ai.client.operation.duration`` for failed
operations as well as successful ones, with ``error.type`` set to the class of the error.
Recording only successes leaves error latency out of the metric and gives no error rate.
"""
if operation_duration_histogram is None or duration is None:
return
attrs = _filter_metric_attributes(attributes)
attrs[OtelAttr.ERROR_TYPE] = type(exception).__name__
operation_duration_histogram.record(duration, attributes=attrs)


def _capture_response(
span: trace.Span,
attributes: dict[str, Any],
Expand All @@ -3526,7 +3588,7 @@ def _capture_response(
) -> None:
"""Set the response for a given span."""
span.set_attributes(attributes)
attrs: dict[str, Any] = {k: v for k, v in attributes.items() if k in GEN_AI_METRIC_ATTRIBUTES}
attrs = _filter_metric_attributes(attributes)
if token_usage_histogram and (input_tokens := attributes.get(OtelAttr.INPUT_TOKENS)) is not None:
token_usage_histogram.record(input_tokens, attributes={**attrs, OtelAttr.T_TYPE: OtelAttr.T_TYPE_INPUT})
if token_usage_histogram and (output_tokens := attributes.get(OtelAttr.OUTPUT_TOKENS)) is not None:
Expand Down
Loading
Loading