scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-13 23:11:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 100

Merge pull request #35277 from BerriAI/litellm_/backport-35271-1-94-x-ba1512

Browse source 
chore(release): backport #35271 to stable/1.94.x and cut 1.94.1
This commit is contained in:
yuneng-jiang 2026-07-30 15:51:10 -07:00 • committed by GitHub
commit 065bf83f0c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 46 additions and 176 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
litellm/proxy/_types.py
View file

@ -2433,16 +2433,6 @@ class ConfigGeneralSettings(LiteLLMPydanticObjectBase):
"is active as a reminder that hard enforcement is relaxed."
),
)
skip_user_budget_on_team_key: bool | None = Field(
None,
description=(
"If True, restores the legacy behavior where a user's personal "
"max_budget is NOT enforced when their key belongs to a team; only "
"the team (and team-member) budgets apply. Defaults to False, meaning "
"the user's personal max_budget is always enforced regardless of "
"whether the key belongs to a team (see GitHub issue #12905)."
),
)
user_url_validation: Optional[bool] = Field(
None,
description=(

41
litellm/proxy/auth/auth_checks.py
View file

@ -631,31 +631,28 @@ async def common_checks(
)
async def _user_max_budget_check() -> None:
if user_object is None or user_object.max_budget is None:
return
skip_for_team = (
general_settings.get("skip_user_budget_on_team_key") is True
and team_object is not None
and team_object.team_id is not None
)
if skip_for_team:
return
from litellm.proxy.proxy_server import get_current_spend
# 4.1 personal budget, if personal key
if (
(team_object is None or team_object.team_id is None)
and user_object is not None
and user_object.max_budget is not None
):
from litellm.proxy.proxy_server import get_current_spend
user_budget = user_object.max_budget
user_spend = await get_current_spend(
counter_key=f"spend:user:{user_object.user_id}",
fallback_spend=user_object.spend or 0.0,
max_budget=user_budget,
)
if math.isfinite(user_budget) and user_spend >= user_budget:
raise litellm.BudgetExceededError(
current_cost=user_spend,
user_budget = user_object.max_budget
user_spend = await get_current_spend(
counter_key=f"spend:user:{user_object.user_id}",
fallback_spend=user_object.spend or 0.0,
max_budget=user_budget,
message=f"ExceededBudget: User={user_object.user_id} over budget. Spend={user_spend}, Budget={user_budget}",
entity_type=Litellm_EntityType.USER.value,
entity_id=user_object.user_id,
)
if math.isfinite(user_budget) and user_spend >= user_budget:
raise litellm.BudgetExceededError(
current_cost=user_spend,
max_budget=user_budget,
message=f"ExceededBudget: User={user_object.user_id} over budget. Spend={user_spend}, Budget={user_budget}",
entity_type=Litellm_EntityType.USER.value,
entity_id=user_object.user_id,
)
# Each scope reads a distinct counter key with no cross-scope ordering
# dependency, so the per-scope Redis-first reads run concurrently instead

1
litellm/proxy/auth/user_api_key_auth.py
View file

@ -2439,7 +2439,6 @@ async def _reserve_budget_after_common_checks(
proxy_logging_obj=proxy_logging_obj,
end_user_id=end_user_id,
end_user_object=end_user_object,
skip_user_budget_on_team_key=general_settings.get("skip_user_budget_on_team_key") is True,
)

1
litellm/proxy/proxy_server.py
View file

@ -15016,7 +15016,6 @@ async def get_config_list(
"forward_client_headers_to_llm_api": {"type": "Boolean"},
"mcp_required_fields": {"type": "List"},
"cancel_on_disconnect": {"type": "Boolean"},
"skip_user_budget_on_team_key": {"type": "Boolean"},
"disable_auto_add_proxy_admin_to_teams": {"type": "Boolean"},
}

6
litellm/proxy/spend_tracking/budget_reservation.py
View file

@ -137,7 +137,6 @@ async def reserve_budget_for_request(
proxy_logging_obj: ProxyLogging,
end_user_id: Optional[str] = None,
end_user_object: Optional[Any] = None,
skip_user_budget_on_team_key: bool = False,
) -> Optional[dict]:
if valid_token is None or not RouteChecks.is_llm_api_route(route=route):
return None
@ -156,7 +155,6 @@ async def reserve_budget_for_request(
proxy_logging_obj=proxy_logging_obj,
end_user_id=end_user_id,
end_user_object=end_user_object,
skip_user_budget_on_team_key=skip_user_budget_on_team_key,
)
if not counters:
return None
@ -312,7 +310,6 @@ async def _get_budget_counters(
proxy_logging_obj: ProxyLogging,
end_user_id: Optional[str] = None,
end_user_object: Optional[Any] = None,
skip_user_budget_on_team_key: bool = False,
) -> List[_BudgetCounter]:
counters: List[_BudgetCounter] = []
@ -361,9 +358,8 @@ async def _get_budget_counters(
)
)
is_team_key = team_object is not None and team_object.team_id is not None
if (
not (is_team_key and skip_user_budget_on_team_key)
(team_object is None or team_object.team_id is None)
and user_object is not None
and user_object.user_id is not None
and user_object.max_budget is not None

4
pyproject.toml
View file

@ -1,6 +1,6 @@
[project]
name = "litellm"
version = "1.94.0"
version = "1.94.1"
description = "Library to easily interface with LLM API providers"
readme = "README.md"
requires-python = ">=3.10, <3.15"
@ -289,7 +289,7 @@ members = ["enterprise", "litellm-proxy-extras"]
profile = "black"
[tool.commitizen]
version = "1.94.0"
version = "1.94.1"
version_files = [
"pyproject.toml:^version",
]

11
tests/proxy_unit_tests/test_user_api_key_auth.py
View file

@ -219,8 +219,8 @@ async def test_aaauser_personal_budgets(key_ownership):
"""
Set a personal budget on a user
User budget is enforced regardless of key ownership (personal or team).
Both cases should raise BudgetExceededError when the user is over budget.
- have it only apply when key belongs to user -> raises BudgetExceededError
- if key belongs to team, have key respect team budget -> allows call to go through
"""
import asyncio
import time
@ -278,9 +278,12 @@ async def test_aaauser_personal_budgets(key_ownership):
== valid_token
)
with pytest.raises(ProxyException) as exc_info:
if key_ownership == "user_key":
with pytest.raises(ProxyException) as exc_info:
await user_api_key_auth(request=request, api_key="Bearer " + user_key)
assert exc_info.value.type == ProxyErrorTypes.budget_exceeded
else:
await user_api_key_auth(request=request, api_key="Bearer " + user_key)
assert exc_info.value.type == ProxyErrorTypes.budget_exceeded
@pytest.mark.asyncio

58
tests/test_litellm/proxy/auth/test_auth_checks.py
View file

@ -4583,66 +4583,26 @@ async def test_common_checks_personal_user_budget_blocks_in_gather():
@pytest.mark.asyncio
async def test_user_budget_enforced_on_team_key():
"""User budget must be enforced even when the key belongs to a team.
async def test_common_checks_personal_user_budget_skipped_for_team_key():
"""A user's personal max_budget does not apply to a team-scoped key.
Previously _user_max_budget_check skipped enforcement for team keys,
letting a user with a $100 personal budget spend unlimited through a
team key. This regression test ensures that is no longer the case.
Team keys are governed by the team (and team-member) budgets only; the key
owner's personal budget is deliberately out of scope. This asserts the read
path lets a team key through even when the user is far over their personal
budget, and fails if personal enforcement is reintroduced for team keys.
"""
from fastapi import Request
from litellm.proxy.auth.auth_checks import common_checks
user = LiteLLM_UserTable(user_id="u1", spend=0.0, max_budget=100.0)
team = LiteLLM_TeamTable(team_id="t1", max_budget=2100.0)
team = LiteLLM_TeamTable(team_id="t1", spend=0.0, max_budget=1000.0)
token = UserAPIKeyAuth(token="k1", user_id="u1", team_id="t1")
async def _spend_by_counter(counter_key, fallback_spend, max_budget=None, **kwargs):
return 999.0 if counter_key == "spend:user:u1" else 0.0
async def _no_membership(*a, **kw):
return None
with patch("litellm.proxy.proxy_server.prisma_client", None), patch(
"litellm.proxy.proxy_server.get_current_spend", _spend_by_counter
), patch("litellm.proxy.auth.auth_checks.get_team_membership", _no_membership):
with pytest.raises(litellm.BudgetExceededError) as over:
await common_checks(
request_body={"messages": [{"role": "user", "content": "hi"}]},
team_object=team,
user_object=user,
end_user_object=None,
global_proxy_spend=None,
general_settings={},
route="/chat/completions",
llm_router=None,
proxy_logging_obj=MagicMock(),
valid_token=token,
request=MagicMock(spec=Request),
)
assert "User=u1" in str(over.value)
@pytest.mark.asyncio
async def test_skip_user_budget_on_team_key_flag_restores_old_behavior():
"""Setting skip_user_budget_on_team_key=True skips user budget for team keys.
This is the opt-in escape hatch that restores the legacy behavior where
user budgets were not enforced when the key belonged to a team.
"""
from fastapi import Request
from litellm.proxy.auth.auth_checks import common_checks
user = LiteLLM_UserTable(user_id="u1", spend=0.0, max_budget=100.0)
team = LiteLLM_TeamTable(team_id="t1", max_budget=2100.0)
token = UserAPIKeyAuth(token="k1", user_id="u1", team_id="t1")
async def _spend_by_counter(counter_key, fallback_spend, max_budget=None, **kwargs):
return 999.0 if counter_key == "spend:user:u1" else 0.0
async def _no_membership(*a, **kw):
async def _no_membership(*args, **kwargs):
return None
with patch("litellm.proxy.proxy_server.prisma_client", None), patch(
@ -4654,7 +4614,7 @@ async def test_skip_user_budget_on_team_key_flag_restores_old_behavior():
user_object=user,
end_user_object=None,
global_proxy_spend=None,
general_settings={"skip_user_budget_on_team_key": True},
general_settings={},
route="/chat/completions",
llm_router=None,
proxy_logging_obj=MagicMock(),

49
tests/test_litellm/proxy/test_budget_reservation.py
View file

@ -610,12 +610,12 @@ async def test_should_reserve_team_member_and_org_budget_counters(spend_counter_
@pytest.mark.asyncio
async def test_should_reserve_user_budget_counter_for_team_key(spend_counter_state):
"""A user's personal budget must be reserved even when the key belongs to a team.
async def test_should_not_reserve_user_budget_counter_for_team_key(spend_counter_state):
"""The reservation path mirrors the read path: no personal user counter for a team key.
Regression for GitHub issue #12905: previously the reservation path skipped the
user spend counter whenever the key had a team, so a team key could overshoot the
user's personal max_budget under concurrency.
A team-scoped key reserves against the key and team counters only, so the key
owner's personal max_budget never gates a team request. Fails if the user
counter is reserved for team keys again.
"""
counter_cache, key_cache = spend_counter_state
proxy_logging_obj = ProxyLogging(user_api_key_cache=key_cache)
@ -644,44 +644,7 @@ async def test_should_reserve_user_budget_counter_for_team_key(spend_counter_sta
proxy_logging_obj=proxy_logging_obj,
)
assert counter_cache.in_memory_cache.get_cache(key="spend:user:user-on-team") == pytest.approx(0.3)
await release_budget_reservation(reservation)
@pytest.mark.asyncio
async def test_should_skip_user_budget_counter_for_team_key_when_flag_set(spend_counter_state):
"""skip_user_budget_on_team_key=True restores the legacy behavior where a user's
personal budget is not reserved for a team key."""
counter_cache, key_cache = spend_counter_state
proxy_logging_obj = ProxyLogging(user_api_key_cache=key_cache)
valid_token = UserAPIKeyAuth(
token="key-user-on-team-skip",
spend=0.0,
user_id="user-on-team-skip",
team_id="team-no-budget-skip",
)
team_object = LiteLLM_TeamTable(team_id="team-no-budget-skip", spend=0.0, max_budget=None)
user_object = LiteLLM_UserTable(user_id="user-on-team-skip", spend=0.0, max_budget=5.0)
with patch(
"litellm.proxy.spend_tracking.budget_reservation.estimate_request_max_cost",
return_value=0.3,
):
reservation = await reserve_budget_for_request(
request_body=_request_body(),
route="/chat/completions",
llm_router=None,
valid_token=valid_token,
team_object=team_object,
user_object=user_object,
prisma_client=None,
user_api_key_cache=key_cache,
proxy_logging_obj=proxy_logging_obj,
skip_user_budget_on_team_key=True,
)
assert counter_cache.in_memory_cache.get_cache(key="spend:user:user-on-team-skip") is None
assert counter_cache.in_memory_cache.get_cache(key="spend:user:user-on-team") is None
await release_budget_reservation(reservation)

32
tests/test_litellm/proxy/test_proxy_server.py
View file

@ -8875,38 +8875,6 @@ def test_get_config_list_includes_cancel_on_disconnect(monkeypatch):
app.dependency_overrides.clear()
def test_get_config_list_includes_skip_user_budget_on_team_key(monkeypatch):
"""Related to #12905: the opt-out flag must be discoverable via /config/list so
it renders as a Boolean toggle on the Admin UI General Settings table. This
requires both the ConfigGeneralSettings field and the allowed_args entry."""
import types
from unittest.mock import AsyncMock, MagicMock
from fastapi.testclient import TestClient
import litellm.proxy.proxy_server as ps
from litellm.proxy._types import LitellmUserRoles, UserAPIKeyAuth
from litellm.proxy.proxy_server import app
mock_prisma = MagicMock()
mock_config_table = MagicMock()
mock_config_table.find_first = AsyncMock(return_value=None)
mock_prisma.db = types.SimpleNamespace(litellm_config=mock_config_table)
monkeypatch.setattr(ps, "prisma_client", mock_prisma)
app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth(
user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN
)
try:
client = TestClient(app)
resp = client.get("/config/list", params={"config_type": "general_settings"})
assert resp.status_code == 200, resp.text
fields = {item["field_name"]: item for item in resp.json()}
assert "skip_user_budget_on_team_key" in fields
assert fields["skip_user_budget_on_team_key"]["field_type"] == "Boolean"
finally:
app.dependency_overrides.clear()
def test_get_config_list_includes_budget_exceeded_throttle_percentage(monkeypatch):
"""The throttle fraction is a litellm_settings scalar surfaced on the General
Settings table as a Float field so it sits with the other global limits; it

5
ui/litellm-dashboard/src/lib/http/schema.d.ts generated vendored
View file

@ -22666,11 +22666,6 @@ export interface components {
* @description When set to True, rejects requests that contain client-side 'metadata.tags' to prevent users from influencing budgets by sending different tags. Tags can only be inherited from the API key metadata.
*/
reject_clientside_metadata_tags?: boolean | null;
/**
* Skip User Budget On Team Key
* @description If True, restores the legacy behavior where a user's personal max_budget is NOT enforced when their key belongs to a team; only the team (and team-member) budgets apply. Defaults to False, meaning the user's personal max_budget is always enforced regardless of whether the key belongs to a team (see GitHub issue #12905).
*/
skip_user_budget_on_team_key?: boolean | null;
/**
* Store Model In Db
* @description If True, models and config are stored in and loaded from the database. Default is False.

4
uv.lock generated
View file

@ -10,7 +10,7 @@ resolution-markers = [
]
[options]
exclude-newer = "2026-07-25T00:00:51.267256Z"
exclude-newer = "2026-07-27T20:58:05.848413Z"
exclude-newer-span = "P3D"
[manifest]
@ -3966,7 +3966,7 @@ wheels = [
[[package]]
name = "litellm"
version = "1.94.0"
version = "1.94.1"
source = { editable = "." }
dependencies = [
{ name = "aiohttp" },