llama-stack-mirror/llama_stack/providers/utils/telemetry/dataset_mixin.py
Dinesh Yeduguru c23363d561
Add ability to query and export spans to dataset (#574)
This PR adds two new methods to the telemetry API:
1) Gives the ability to query spans directly instead of first querying
traces and then using that to get spans
2) Another method save_spans_to_dataset, which builds on the query spans
to save it on dataset.

This give the ability to saves spans that are part of an agent session
to a dataset.

The unique aspect of this API is that we dont require each provider of
telemetry to implement this method. Hence, its implemented in the
protocol class itself. This required the protocol check to be slightly
modified.
2024-12-05 21:07:30 -08:00

87 lines
2.9 KiB
Python

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from typing import List, Optional
from llama_stack.apis.datasetio import DatasetIO
from llama_stack.apis.telemetry import QueryCondition, Span, SpanWithChildren
class TelemetryDatasetMixin:
"""Mixin class that provides dataset-related functionality for telemetry providers."""
datasetio_api: DatasetIO
async def save_spans_to_dataset(
self,
attribute_filters: List[QueryCondition],
attributes_to_save: List[str],
dataset_id: str,
max_depth: Optional[int] = None,
) -> None:
spans = await self.query_spans(
attribute_filters=attribute_filters,
attributes_to_return=attributes_to_save,
max_depth=max_depth,
)
rows = [
{
"trace_id": span.trace_id,
"span_id": span.span_id,
"parent_span_id": span.parent_span_id,
"name": span.name,
"start_time": span.start_time,
"end_time": span.end_time,
**{attr: span.attributes.get(attr) for attr in attributes_to_save},
}
for span in spans
]
await self.datasetio_api.append_rows(dataset_id=dataset_id, rows=rows)
async def query_spans(
self,
attribute_filters: List[QueryCondition],
attributes_to_return: List[str],
max_depth: Optional[int] = None,
) -> List[Span]:
traces = await self.query_traces(attribute_filters=attribute_filters)
spans = []
for trace in traces:
span_tree = await self.get_span_tree(
span_id=trace.root_span_id,
attributes_to_return=attributes_to_return,
max_depth=max_depth,
)
def extract_spans(span: SpanWithChildren) -> List[Span]:
result = []
if span.attributes and all(
attr in span.attributes and span.attributes[attr] is not None
for attr in attributes_to_return
):
result.append(
Span(
trace_id=trace.root_span_id,
span_id=span.span_id,
parent_span_id=span.parent_span_id,
name=span.name,
start_time=span.start_time,
end_time=span.end_time,
attributes=span.attributes,
)
)
for child in span.children:
result.extend(extract_spans(child))
return result
spans.extend(extract_spans(span_tree))
return spans