llama-stack/llama_stack/providers/utils/telemetry/sqlite_trace_store.py
Dinesh Yeduguru fcd6449519
Telemetry API redesign (#525)
# What does this PR do?
Change the Telemetry API to be able to support different use cases like
returning traces for the UI and ability to export for Evals.
Other changes:
* Add a new trace_protocol decorator to decorate all our API methods so
that any call to them will automatically get traced across all impls.
* There is some issue with the decorator pattern of span creation when
using async generators, where there are multiple yields with in the same
context. I think its much more explicit by using the explicit context
manager pattern using with. I moved the span creations in agent instance
to be using with
* Inject session id at the turn level, which should quickly give us all
traces across turns for a given session

Addresses #509

## Test Plan
```
llama stack run /Users/dineshyv/.llama/distributions/llamastack-together/together-run.yaml
PYTHONPATH=. python -m examples.agents.rag_with_memory_bank localhost 5000


 curl -X POST 'http://localhost:5000/alpha/telemetry/query-traces' \
-H 'Content-Type: application/json' \
-d '{
  "attribute_filters": [
    {
      "key": "session_id",
      "op": "eq",
      "value": "dd667b87-ca4b-4d30-9265-5a0de318fc65" }],
  "limit": 100,
  "offset": 0,
  "order_by": ["start_time"]
}' | jq .
[
  {
    "trace_id": "6902f54b83b4b48be18a6f422b13e16f",
    "root_span_id": "5f37b85543afc15a",
    "start_time": "2024-12-04T08:08:30.501587",
    "end_time": "2024-12-04T08:08:36.026463"
  },
  {
    "trace_id": "92227dac84c0615ed741be393813fb5f",
    "root_span_id": "af7c5bb46665c2c8",
    "start_time": "2024-12-04T08:08:36.031170",
    "end_time": "2024-12-04T08:08:41.693301"
  },
  {
    "trace_id": "7d578a6edac62f204ab479fba82f77b6",
    "root_span_id": "1d935e3362676896",
    "start_time": "2024-12-04T08:08:41.695204",
    "end_time": "2024-12-04T08:08:47.228016"
  },
  {
    "trace_id": "dbd767d76991bc816f9f078907dc9ff2",
    "root_span_id": "f5a7ee76683b9602",
    "start_time": "2024-12-04T08:08:47.234578",
    "end_time": "2024-12-04T08:08:53.189412"
  }
]


curl -X POST 'http://localhost:5000/alpha/telemetry/get-span-tree' \
-H 'Content-Type: application/json' \
-d '{ "span_id" : "6cceb4b48a156913", "max_depth": 2, "attributes_to_return": ["input"] }' | jq .
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   875  100   790  100    85  18462   1986 --:--:-- --:--:-- --:--:-- 20833
{
  "span_id": "6cceb4b48a156913",
  "trace_id": "dafa796f6aaf925f511c04cd7c67fdda",
  "parent_span_id": "892a66d726c7f990",
  "name": "retrieve_rag_context",
  "start_time": "2024-12-04T09:28:21.781995",
  "end_time": "2024-12-04T09:28:21.913352",
  "attributes": {
    "input": [
      "{\"role\":\"system\",\"content\":\"You are a helpful assistant\"}",
      "{\"role\":\"user\",\"content\":\"What are the top 5 topics that were explained in the documentation? Only list succinct bullet points.\",\"context\":null}"
    ]
  },
  "children": [
    {
      "span_id": "1a2df181854064a8",
      "trace_id": "dafa796f6aaf925f511c04cd7c67fdda",
      "parent_span_id": "6cceb4b48a156913",
      "name": "MemoryRouter.query_documents",
      "start_time": "2024-12-04T09:28:21.787620",
      "end_time": "2024-12-04T09:28:21.906512",
      "attributes": {
        "input": null
      },
      "children": [],
      "status": "ok"
    }
  ],
  "status": "ok"
}

```

<img width="1677" alt="Screenshot 2024-12-04 at 9 42 56 AM"
src="https://github.com/user-attachments/assets/4d3cea93-05ce-415a-93d9-4b1628631bf8">
2024-12-04 11:22:45 -08:00

180 lines
6.2 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.
import json
from datetime import datetime
from typing import List, Optional, Protocol
import aiosqlite
from llama_stack.apis.telemetry import QueryCondition, SpanWithChildren, Trace
class TraceStore(Protocol):
async def query_traces(
self,
attribute_filters: Optional[List[QueryCondition]] = None,
limit: Optional[int] = 100,
offset: Optional[int] = 0,
order_by: Optional[List[str]] = None,
) -> List[Trace]: ...
async def get_materialized_span(
self,
span_id: str,
attributes_to_return: Optional[List[str]] = None,
max_depth: Optional[int] = None,
) -> SpanWithChildren: ...
class SQLiteTraceStore(TraceStore):
def __init__(self, conn_string: str):
self.conn_string = conn_string
async def query_traces(
self,
attribute_filters: Optional[List[QueryCondition]] = None,
limit: Optional[int] = 100,
offset: Optional[int] = 0,
order_by: Optional[List[str]] = None,
) -> List[Trace]:
def build_where_clause() -> tuple[str, list]:
if not attribute_filters:
return "", []
ops_map = {"eq": "=", "ne": "!=", "gt": ">", "lt": "<"}
conditions = [
f"json_extract(s.attributes, '$.{condition.key}') {ops_map[condition.op]} ?"
for condition in attribute_filters
]
params = [condition.value for condition in attribute_filters]
where_clause = " WHERE " + " AND ".join(conditions)
return where_clause, params
def build_order_clause() -> str:
if not order_by:
return ""
order_clauses = []
for field in order_by:
desc = field.startswith("-")
clean_field = field[1:] if desc else field
order_clauses.append(f"t.{clean_field} {'DESC' if desc else 'ASC'}")
return " ORDER BY " + ", ".join(order_clauses)
# Build the main query
base_query = """
WITH matching_traces AS (
SELECT DISTINCT t.trace_id
FROM traces t
JOIN spans s ON t.trace_id = s.trace_id
{where_clause}
),
filtered_traces AS (
SELECT t.trace_id, t.root_span_id, t.start_time, t.end_time
FROM matching_traces mt
JOIN traces t ON mt.trace_id = t.trace_id
LEFT JOIN spans s ON t.trace_id = s.trace_id
{order_clause}
)
SELECT DISTINCT trace_id, root_span_id, start_time, end_time
FROM filtered_traces
LIMIT {limit} OFFSET {offset}
"""
where_clause, params = build_where_clause()
query = base_query.format(
where_clause=where_clause,
order_clause=build_order_clause(),
limit=limit,
offset=offset,
)
# Execute query and return results
async with aiosqlite.connect(self.conn_string) as conn:
conn.row_factory = aiosqlite.Row
async with conn.execute(query, params) as cursor:
rows = await cursor.fetchall()
return [
Trace(
trace_id=row["trace_id"],
root_span_id=row["root_span_id"],
start_time=datetime.fromisoformat(row["start_time"]),
end_time=datetime.fromisoformat(row["end_time"]),
)
for row in rows
]
async def get_materialized_span(
self,
span_id: str,
attributes_to_return: Optional[List[str]] = None,
max_depth: Optional[int] = None,
) -> SpanWithChildren:
# Build the attributes selection
attributes_select = "s.attributes"
if attributes_to_return:
json_object = ", ".join(
f"'{key}', json_extract(s.attributes, '$.{key}')"
for key in attributes_to_return
)
attributes_select = f"json_object({json_object})"
# SQLite CTE query with filtered attributes
query = f"""
WITH RECURSIVE span_tree AS (
SELECT s.*, 1 as depth, {attributes_select} as filtered_attributes
FROM spans s
WHERE s.span_id = ?
UNION ALL
SELECT s.*, st.depth + 1, {attributes_select} as filtered_attributes
FROM spans s
JOIN span_tree st ON s.parent_span_id = st.span_id
WHERE (? IS NULL OR st.depth < ?)
)
SELECT *
FROM span_tree
ORDER BY depth, start_time
"""
async with aiosqlite.connect(self.conn_string) as conn:
conn.row_factory = aiosqlite.Row
async with conn.execute(query, (span_id, max_depth, max_depth)) as cursor:
rows = await cursor.fetchall()
if not rows:
raise ValueError(f"Span {span_id} not found")
# Build span tree
spans_by_id = {}
root_span = None
for row in rows:
span = SpanWithChildren(
span_id=row["span_id"],
trace_id=row["trace_id"],
parent_span_id=row["parent_span_id"],
name=row["name"],
start_time=datetime.fromisoformat(row["start_time"]),
end_time=datetime.fromisoformat(row["end_time"]),
attributes=json.loads(row["filtered_attributes"]),
status=row["status"].lower(),
children=[],
)
spans_by_id[span.span_id] = span
if span.span_id == span_id:
root_span = span
elif span.parent_span_id in spans_by_id:
spans_by_id[span.parent_span_id].children.append(span)
return root_span