From cf3af0f486590633ee875f6d120c416509fad5d2 Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 3 Sep 2026 23:53:59 +0000 Subject: [PATCH] perf(spend): group /spend/logs summary by day in Postgres instead of per-row Prisma group_by (#39351) * perf(spend): group /spend/logs summary by day in Postgres instead of per-row Prisma group_by Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * refactor(spend): simplify /spend/logs daily summary aggregation Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend): preserve spend logs response schema Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(spend): compare spend log range bounds as naive UTC timestamps Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * test: cover spend logs summary edge cases Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * test: cover spend summary request filters Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * chore: ratchet lint budgets after merge Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: yassin Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- basedpyright-code-budget.json | 10 +- .../spend_management_endpoints.py | 164 +++++++----- ruff-strict-budget.json | 2 +- .../test_spend_management_endpoints.py | 235 +++++++++++++++--- type-discipline-budget.json | 6 +- 5 files changed, 303 insertions(+), 114 deletions(-) diff --git a/basedpyright-code-budget.json b/basedpyright-code-budget.json index 9a9b1138a7d..cb3756575f7 100644 --- a/basedpyright-code-budget.json +++ b/basedpyright-code-budget.json @@ -3,7 +3,7 @@ "limit": 14074 }, "reportArgumentType": { - "limit": 2214 + "limit": 2206 }, "reportAssignmentType": { "limit": 319 @@ -57,7 +57,7 @@ "limit": 5601 }, "reportMissingTypeArgument": { - "limit": 15287 + "limit": 15285 }, "reportMissingTypeStubs": { "limit": 40 @@ -99,19 +99,19 @@ "limit": 0 }, "reportUnknownArgumentType": { - "limit": 44362 + "limit": 44360 }, "reportUnknownLambdaType": { "limit": 109 }, "reportUnknownMemberType": { - "limit": 38323 + "limit": 38311 }, "reportUnknownParameterType": { "limit": 19624 }, "reportUnknownVariableType": { - "limit": 29861 + "limit": 29847 }, "reportUnnecessaryCast": { "limit": 111 diff --git a/litellm/proxy/spend_tracking/spend_management_endpoints.py b/litellm/proxy/spend_tracking/spend_management_endpoints.py index b86a877e8f9..dff100bdea7 100644 --- a/litellm/proxy/spend_tracking/spend_management_endpoints.py +++ b/litellm/proxy/spend_tracking/spend_management_endpoints.py @@ -3,7 +3,8 @@ import collections import json import os from collections.abc import Mapping, Sequence -from datetime import datetime, timedelta, timezone +from datetime import date, datetime, timedelta, timezone +from itertools import groupby from types import MappingProxyType from typing import ( TYPE_CHECKING, @@ -16,7 +17,6 @@ from typing import ( TypeAlias, TypedDict, TypeVar, - cast, # noqa: TID251 # prisma group_by returns untyped aggregate mappings ) import fastapi @@ -201,16 +201,12 @@ class _SessionSpendStats(NamedTuple): _SessionSpendMap: TypeAlias = Mapping[tuple[str, str], _SessionSpendStats] -class _SpendSumAggregate(TypedDict, total=False): - spend: ReadOnly[float] - - -class _SpendGroupByRow(TypedDict): +class _SpendDailySummaryRow(TypedDict): + day: ReadOnly[str] api_key: ReadOnly[str] user: ReadOnly[str | None] model: ReadOnly[str] - startTime: ReadOnly[object] - _sum: ReadOnly[_SpendSumAggregate] + spend: ReadOnly[float] async def _query_raw(prisma_client: PrismaClient, sql_query: str, *args: object) -> Sequence[_RowT]: @@ -251,6 +247,66 @@ def _verification_token_table(prisma_client: PrismaClient) -> _VerificationToken return VerificationTokenRepository(prisma_client).table +def _spend_logs_daily_summary_sql( + *, + start_date_iso: str, + end_date_iso: str, + api_key: str | None, + request_id: str | None, + user_id: str | None, +) -> tuple[str, tuple[object, ...]]: + filter_params: Final[tuple[tuple[str, object], ...]] = tuple( + (column, value) + for column, value in ( + ("api_key", api_key), + ("request_id", request_id), + ('"user"', user_id), + ) + if value is not None + ) + filter_clauses: Final[tuple[str, ...]] = tuple( + f"AND {column} = ${index}" for index, (column, _) in enumerate(filter_params, start=3) + ) + filter_sql: Final = "\n".join(filter_clauses) + sql_query: Final = f""" +SELECT + to_char(date_trunc('day', "startTime"), 'YYYY-MM-DD') AS day, + api_key, + "user", + model, + SUM(spend) AS spend +FROM "LiteLLM_SpendLogs" +WHERE "startTime" >= ($1::timestamptz AT TIME ZONE 'UTC') AND "startTime" <= ($2::timestamptz AT TIME ZONE 'UTC') +{filter_sql} +GROUP BY 1, 2, 3, 4 +ORDER BY 1 +""" + params: Final[tuple[object, ...]] = ( + start_date_iso, + end_date_iso, + *(value for _, value in filter_params), + ) + return sql_query, params + + +def _sum_spend_by( + rows: Sequence[_SpendDailySummaryRow], column: Literal["api_key", "user", "model"] +) -> Mapping[str | None, float]: + keys: Final = frozenset(row[column] for row in rows) + return {key: sum(float(row["spend"]) for row in rows if row[column] == key) for key in keys} + + +def _daily_summary_item(summary_date: date, rows: Sequence[_SpendDailySummaryRow]) -> Mapping[str, object]: + api_key_spend: Final = {key: value for key, value in _sum_spend_by(rows, "api_key").items() if key is not None} + return { + **api_key_spend, + "startTime": summary_date, + "spend": sum(float(row["spend"]) for row in rows), + "users": _sum_spend_by(rows, "user"), + "models": _sum_spend_by(rows, "model"), + } + + async def _find_spend_logs( prisma_client: PrismaClient, where: Mapping[str, object], @@ -3266,18 +3322,22 @@ async def view_spend_logs( start_date_iso: Final = start_date_obj.isoformat() end_date_iso: Final = end_date_obj.isoformat() - filter_query: Final = { + filter_query: Final[ + dict[str, object] + ] = { # mutable-ok: legacy filters are extended for optional parameters "startTime": { "gte": start_date_iso, # Greater than or equal to Start Date "lte": end_date_iso, # Less than or equal to End Date } } + summary_api_key: Final[str | None] = ( + prisma_client.hash_token(token=api_key) + if api_key is not None and api_key.startswith("sk-") + else api_key + ) if api_key is not None and isinstance(api_key, str): - if api_key.startswith("sk-"): - filter_query["api_key"] = prisma_client.hash_token(token=api_key) - else: - filter_query["api_key"] = api_key + filter_query["api_key"] = summary_api_key if request_id is not None and isinstance(request_id, str): filter_query["request_id"] = request_id if user_id is not None and isinstance(user_id, str): @@ -3296,58 +3356,34 @@ async def view_spend_logs( return data # Legacy behavior: return summarized data (when summarize=true) - # SQL query - response: Final = await SpendLogsRepository(prisma_client).table.group_by( - by=["api_key", "user", "model", "startTime"], - where=filter_query, - sum={ - "spend": True, - }, + summary_sql_and_params: Final = _spend_logs_daily_summary_sql( + start_date_iso=start_date_iso, + end_date_iso=end_date_iso, + api_key=summary_api_key, + request_id=request_id, + user_id=user_id, ) + sql_query, params = summary_sql_and_params + rows: Final[Sequence[_SpendDailySummaryRow]] = await _query_raw(prisma_client, sql_query, *params) + if len(rows) == 0: + return [] # pyright: ignore[reportUnknownVariableType] # empty summary has no element type - if isinstance(response, list) and len(response) > 0 and isinstance(response[0], dict): - spend_rows: Final = cast(Sequence[_SpendGroupByRow], response) # cast-ok: by/sum fix the shape - result: Final[dict] = {} - for record in spend_rows: - dt_object = datetime.strptime(str(record["startTime"]), "%Y-%m-%dT%H:%M:%S.%fZ") - date = dt_object.date() - if date not in result: - result[date] = {"users": {}, "models": {}} - api_key = record["api_key"] - user_id = record["user"] - model = record["model"] - result[date]["spend"] = result[date].get("spend", 0) + record.get("_sum", {}).get("spend", 0) - result[date][api_key] = result[date].get(api_key, 0) + record.get("_sum", {}).get("spend", 0) - result[date]["users"][user_id] = result[date]["users"].get(user_id, 0) + record.get("_sum", {}).get( - "spend", 0 - ) - result[date]["models"][model] = result[date]["models"].get(model, 0) + record.get("_sum", {}).get( - "spend", 0 - ) - return_list: Final = [] - final_date = None - for k, v in sorted(result.items()): - return_list.append({**v, "startTime": k}) - final_date = k - - end_date_date: Final = end_date_obj.date() - if final_date is not None and final_date < end_date_date: - current_date = final_date + timedelta(days=1) - while current_date <= end_date_date: - # Represent current_date as string because original response has it this way - return_list.append( - { - "startTime": current_date, - "spend": 0, - "users": {}, - "models": {}, - } - ) # If no data, will stay as zero - current_date += timedelta(days=1) # Move on to the next day - - return return_list - - return response + summary_items: Final = tuple( + _daily_summary_item(date.fromisoformat(day), tuple(day_rows)) + for day, day_rows in groupby(rows, key=lambda row: row["day"]) + ) + final_date: Final = date.fromisoformat(rows[-1]["day"]) + end_date_date: Final = end_date_obj.date() + padding: Final[tuple[Mapping[str, object], ...]] = tuple( + { + "startTime": final_date + timedelta(days=offset), + "spend": 0, + "users": {}, + "models": {}, + } + for offset in range(1, (end_date_date - final_date).days + 1) + ) + return [*summary_items, *padding] else: scoped_filter: Final[dict[str, str]] = {} diff --git a/ruff-strict-budget.json b/ruff-strict-budget.json index 24c0ff6b181..8763318b4eb 100644 --- a/ruff-strict-budget.json +++ b/ruff-strict-budget.json @@ -96,7 +96,7 @@ "limit": 10 }, "DTZ007": { - "limit": 17 + "limit": 6 }, "DTZ011": { "limit": 3 diff --git a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py index 73a29afd9b9..329a33eb440 100644 --- a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py +++ b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py @@ -5,14 +5,12 @@ import hashlib import json import re from datetime import timezone +from unittest.mock import AsyncMock, MagicMock, patch import pytest from fastapi import HTTPException from fastapi.testclient import TestClient - -from unittest.mock import AsyncMock, MagicMock, patch - import litellm import litellm.proxy.proxy_server as ps @@ -3325,7 +3323,7 @@ def _compare_nested_dicts( return differences # Check for keys in actual but not in expected - for key in actual.keys(): + for key in actual: current_path = f"{path}.{key}" if path else key if current_path not in ignore_keys and key not in expected: differences.append(f"Extra key in actual: {current_path}") @@ -3495,24 +3493,22 @@ async def test_view_spend_logs_summarize_parameter(client, monkeypatch): # Return individual log entries when summarize=false return mock_spend_logs - async def group_by(self, *args, **kwargs): - # Return grouped data when summarize=true - # Simplified mock response for grouped data + async def query_raw(self, sql_query, *params): yesterday = datetime.datetime.now(timezone.utc) - timedelta(days=1) return [ { "api_key": "sk-test-key", "user": "test_user_1", "model": "gpt-3.5-turbo", - "startTime": yesterday.strftime("%Y-%m-%dT%H:%M:%S.%fZ"), - "_sum": {"spend": 0.05}, + "day": yesterday.date().isoformat(), + "spend": 0.05, }, { "api_key": "sk-test-key", "user": "test_user_1", "model": "gpt-4", - "startTime": yesterday.strftime("%Y-%m-%dT%H:%M:%S.%fZ"), - "_sum": {"spend": 0.10}, + "day": yesterday.date().isoformat(), + "spend": 0.10, }, ] @@ -3850,47 +3846,30 @@ async def test_view_spend_logs_with_date_range_summarized(client, monkeypatch): """ from datetime import datetime, timedelta, timezone - # This simulates the summarized data that Prisma's `group_by` would return. mock_summarized_response = [ { "api_key": "sk-test-key", "user": "test_user_1", "model": "gpt-4", - "startTime": (datetime.now(timezone.utc) - timedelta(days=1)).strftime( - "%Y-%m-%dT%H:%M:%S.%fZ" - ), - "_sum": {"spend": 0.15}, + "day": (datetime.now(timezone.utc) - timedelta(days=1)).date().isoformat(), + "spend": 0.15, } ] - # This mock class will replace the real Prisma client. class MockDB: - def __init__(self): - self.litellm_spendlogs = self - - async def group_by(self, *args, **kwargs): - # We assert that the `gte` and `lte` values are strings in ISO format. - # If they were datetime objects, this test would fail. - where_clause = kwargs.get("where", {}) - start_time_filter = where_clause.get("startTime", {}) - - assert "gte" in start_time_filter - assert "lte" in start_time_filter - assert isinstance(start_time_filter["gte"], str) - assert isinstance(start_time_filter["lte"], str) - assert "T" in start_time_filter["gte"] # Check for ISO format 'T' separator - - # If the assertions pass, return the mock response. + async def query_raw(self, sql_query, *params): + assert isinstance(params[0], str) + assert isinstance(params[1], str) + assert "T" in params[0] + assert "T" in params[1] return mock_summarized_response class MockPrismaClient: def __init__(self): self.db = MockDB() - # Apply the monkeypatch to replace the real prisma_client with our mock. monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", MockPrismaClient()) - # Define a date range for the test. start_date = (datetime.now(timezone.utc) - timedelta(days=2)).strftime("%Y-%m-%d") end_date = datetime.now(timezone.utc).strftime("%Y-%m-%d") @@ -3898,8 +3877,6 @@ async def test_view_spend_logs_with_date_range_summarized(client, monkeypatch): user_role=LitellmUserRoles.PROXY_ADMIN ) try: - # Call the endpoint with both start and end dates. - # We don't need `summarize=true` as it's the default. response = client.get( "/spend/logs", params={ @@ -3909,11 +3886,9 @@ async def test_view_spend_logs_with_date_range_summarized(client, monkeypatch): headers={"Authorization": "Bearer sk-test"}, ) - # ASSERTIONS assert response.status_code == 200 data = response.json() - # Check that the response is not empty and has the summarized structure. assert isinstance(data, list) assert len(data) > 0 assert "startTime" in data[0] @@ -3924,6 +3899,183 @@ async def test_view_spend_logs_with_date_range_summarized(client, monkeypatch): app.dependency_overrides.pop(ps.user_api_key_auth, None) +@pytest.mark.asyncio +async def test_view_spend_logs_summarize_groups_by_day_in_sql(client, monkeypatch): + mock_rows = [ + { + "day": "2024-01-01", + "api_key": "hashed::sk-abc", + "user": "u1", + "model": "gpt-4", + "spend": 0.1, + }, + { + "day": "2024-01-01", + "api_key": "hashed::sk-abc", + "user": "u1", + "model": "gpt-4o", + "spend": 0.2, + }, + ] + + class MockDB: + def __init__(self): + self.captured_sql = None + self.captured_params = None + + async def query_raw(self, sql_query, *params): + self.captured_sql = sql_query + self.captured_params = params + return mock_rows + + class MockPrismaClient: + def __init__(self): + self.db = MockDB() + + def hash_token(self, token): + return "hashed::" + token + + mock_prisma_client = MockPrismaClient() + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", mock_prisma_client) + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN + ) + try: + response = client.get( + "/spend/logs", + params={ + "start_date": "2024-01-01", + "end_date": "2024-01-03", + "api_key": "sk-abc", + "request_id": "req-123", + "user_id": "u1", + }, + headers={"Authorization": "Bearer sk-test"}, + ) + + assert response.status_code == 200 + data = response.json() + sql = mock_prisma_client.db.captured_sql + assert "date_trunc('day'" in sql + assert "GROUP BY" in sql + assert "find_many" not in sql + assert not hasattr(mock_prisma_client.db, "group_by") + assert mock_prisma_client.db.captured_params == ( + "2024-01-01T00:00:00+00:00", + "2024-01-03T00:00:00+00:00", + "hashed::sk-abc", + "req-123", + "u1", + ) + assert len(data) == 3 + assert data[0]["startTime"] == "2024-01-01" + assert data[0]["spend"] == pytest.approx(0.3) + assert data[0]["models"] == {"gpt-4": 0.1, "gpt-4o": 0.2} + assert data[0]["users"] == {"u1": pytest.approx(0.3)} + assert data[0]["hashed::sk-abc"] == pytest.approx(0.3) + assert data[1] == { + "startTime": "2024-01-02", + "spend": 0, + "users": {}, + "models": {}, + } + assert data[2] == { + "startTime": "2024-01-03", + "spend": 0, + "users": {}, + "models": {}, + } + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + +@pytest.mark.asyncio +async def test_view_spend_logs_summarize_empty_rows(client, monkeypatch): + class MockDB: + async def query_raw(self, sql_query, *params): + return [] + + class MockPrismaClient: + def __init__(self): + self.db = MockDB() + + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", MockPrismaClient()) + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN + ) + try: + response = client.get( + "/spend/logs", + params={"start_date": "2024-01-01", "end_date": "2024-01-01"}, + headers={"Authorization": "Bearer sk-test"}, + ) + + assert response.status_code == 200 + assert response.json() == [] + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + +@pytest.mark.asyncio +async def test_view_spend_logs_summarize_unhashed_api_key_without_padding(client, monkeypatch): + mock_rows = [ + { + "day": "2024-01-01", + "api_key": "plain-key", + "user": "u1", + "model": "gpt-4", + "spend": 0.4, + } + ] + + class MockDB: + def __init__(self): + self.captured_params = None + + async def query_raw(self, sql_query, *params): + self.captured_params = params + return mock_rows + + class MockPrismaClient: + def __init__(self): + self.db = MockDB() + + mock_prisma_client = MockPrismaClient() + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", mock_prisma_client) + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN + ) + try: + response = client.get( + "/spend/logs", + params={ + "start_date": "2024-01-01", + "end_date": "2024-01-01", + "api_key": "plain-key", + }, + headers={"Authorization": "Bearer sk-test"}, + ) + + assert response.status_code == 200 + data = response.json() + assert mock_prisma_client.db.captured_params == ( + "2024-01-01T00:00:00+00:00", + "2024-01-01T00:00:00+00:00", + "plain-key", + ) + assert data == [ + { + "startTime": "2024-01-01", + "spend": pytest.approx(0.4), + "plain-key": pytest.approx(0.4), + "users": {"u1": pytest.approx(0.4)}, + "models": {"gpt-4": pytest.approx(0.4)}, + } + ] + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + @pytest.mark.asyncio async def test_ui_view_spend_logs_with_error_code(client): """Test filtering spend logs by error code""" @@ -4832,13 +4984,14 @@ class _CaptureFilterDB: def __init__(self): self.litellm_spendlogs = self self.captured_where = None + self.captured_params = None async def find_many(self, *args, **kwargs): self.captured_where = kwargs.get("where") return [] - async def group_by(self, *args, **kwargs): - self.captured_where = kwargs.get("where") + async def query_raw(self, sql_query, *params): + self.captured_params = params return [] diff --git a/type-discipline-budget.json b/type-discipline-budget.json index 78090779109..ab1a793e09d 100644 --- a/type-discipline-budget.json +++ b/type-discipline-budget.json @@ -3,7 +3,7 @@ "limit": 22328 }, "LIT002": { - "limit": 26758 + "limit": 26750 }, "LIT003": { "limit": 261 @@ -27,10 +27,10 @@ "limit": 0 }, "LIT010": { - "limit": 16470 + "limit": 16468 }, "LIT011": { - "limit": 5516 + "limit": 5514 }, "LIT012": { "limit": 4489