Merge pull request #4139 from BerriAI/litellm_fix_budget_exceeded_error_code

fix(proxy_server.py): use consistent 400-status code error code for exceeded budget errors
This commit is contained in:
Krish Dholakia 2024-06-11 18:36:58 -07:00 committed by GitHub
commit 30d269f93a
6 changed files with 30 additions and 24 deletions

View file

@ -15,7 +15,6 @@ from functools import partial
import dotenv, traceback, random, asyncio, time, contextvars
from copy import deepcopy
import httpx
import litellm
from ._logging import verbose_logger
from litellm import ( # type: ignore

View file

@ -151,8 +151,8 @@ def common_checks(
and route != "/models"
):
if global_proxy_spend > litellm.max_budget:
raise Exception(
f"ExceededBudget: LiteLLM Proxy has exceeded its budget. Current spend: {global_proxy_spend}; Max Budget: {litellm.max_budget}"
raise litellm.BudgetExceededError(
current_cost=global_proxy_spend, max_budget=litellm.max_budget
)
return True

View file

@ -1078,8 +1078,9 @@ async def user_api_key_auth(
_user_id = _user.get("user_id", None)
if user_current_spend > user_max_budget:
raise Exception(
f"ExceededBudget: User {_user_id} has exceeded their budget. Current spend: {user_current_spend}; Max Budget: {user_max_budget}"
raise litellm.BudgetExceededError(
current_cost=user_current_spend,
max_budget=user_max_budget,
)
else:
# Token exists, not expired now check if its in budget for the user
@ -1110,9 +1111,11 @@ async def user_api_key_auth(
)
if user_current_spend > user_max_budget:
raise Exception(
f"ExceededBudget: User {valid_token.user_id} has exceeded their budget. Current spend: {user_current_spend}; Max Budget: {user_max_budget}"
raise litellm.BudgetExceededError(
current_cost=user_current_spend,
max_budget=user_max_budget,
)
# Check 3. Check if user is in their team budget
if valid_token.team_member_spend is not None:
if prisma_client is not None:
@ -1146,8 +1149,9 @@ async def user_api_key_auth(
)
if team_member_budget is not None and team_member_budget > 0:
if valid_token.team_member_spend > team_member_budget:
raise Exception(
f"ExceededBudget: Crossed spend within team. UserID: {valid_token.user_id}, in team {valid_token.team_id} has exceeded their budget. Current spend: {valid_token.team_member_spend}; Max Budget: {team_member_budget}"
raise litellm.BudgetExceededError(
current_cost=valid_token.team_member_spend,
max_budget=team_member_budget,
)
# Check 3. If token is expired
@ -1205,8 +1209,9 @@ async def user_api_key_auth(
####################################
if valid_token.spend >= valid_token.max_budget:
raise Exception(
f"ExceededTokenBudget: Current spend for token: {valid_token.spend}; Max Budget for Token: {valid_token.max_budget}"
raise litellm.BudgetExceededError(
current_cost=valid_token.spend,
max_budget=valid_token.max_budget,
)
# Check 5. Token Model Spend is under Model budget
@ -1242,8 +1247,9 @@ async def user_api_key_auth(
):
current_model_spend = model_spend[0]["_sum"]["spend"]
current_model_budget = max_budget_per_model[current_model]
raise Exception(
f"ExceededModelBudget: Current spend for model: {current_model_spend}; Max Budget for Model: {current_model_budget}"
raise litellm.BudgetExceededError(
current_cost=current_model_spend,
max_budget=current_model_budget,
)
# Check 6. Team spend is under Team budget
@ -1267,8 +1273,9 @@ async def user_api_key_auth(
)
if valid_token.team_spend >= valid_token.team_max_budget:
raise Exception(
f"ExceededTokenBudget: Current Team Spend: {valid_token.team_spend}; Max Budget for Team: {valid_token.team_max_budget}"
raise litellm.BudgetExceededError(
current_cost=valid_token.team_spend,
max_budget=valid_token.team_max_budget,
)
# Check 8: Additional Common Checks across jwt + key auth

View file

@ -471,7 +471,7 @@ def test_call_with_user_over_budget(prisma_client):
asyncio.run(test())
except Exception as e:
error_detail = e.message
assert "Authentication Error, ExceededBudget:" in error_detail
assert "Budget has been exceeded" in error_detail
print(vars(e))
@ -652,7 +652,7 @@ def test_call_with_proxy_over_budget(prisma_client):
error_detail = e.message
else:
error_detail = traceback.format_exc()
assert "Authentication Error, ExceededBudget:" in error_detail
assert "Budget has been exceeded" in error_detail
print(vars(e))
@ -730,7 +730,7 @@ def test_call_with_user_over_budget_stream(prisma_client):
asyncio.run(test())
except Exception as e:
error_detail = e.message
assert "Authentication Error, ExceededBudget:" in error_detail
assert "Budget has been exceeded" in error_detail
print(vars(e))
@ -827,7 +827,7 @@ def test_call_with_proxy_over_budget_stream(prisma_client):
asyncio.run(test())
except Exception as e:
error_detail = e.message
assert "Authentication Error, ExceededBudget:" in error_detail
assert "Budget has been exceeded" in error_detail
print(vars(e))
@ -1362,7 +1362,7 @@ def test_call_with_key_over_budget(prisma_client):
error_detail = e.message
else:
error_detail = str(e)
assert "Authentication Error, ExceededTokenBudget:" in error_detail
assert "Budget has been exceeded" in error_detail
print(vars(e))
@ -1476,7 +1476,7 @@ def test_call_with_key_over_model_budget(prisma_client):
# print(f"Error - {str(e)}")
traceback.print_exc()
error_detail = e.message
assert "Authentication Error, ExceededModelBudget:" in error_detail
assert "Budget has been exceeded!" in error_detail
print(vars(e))
@ -1637,7 +1637,7 @@ async def test_call_with_key_over_budget_stream(prisma_client):
except Exception as e:
print("Got Exception", e)
error_detail = e.message
assert "Authentication Error, ExceededTokenBudget:" in error_detail
assert "Budget has been exceeded" in error_detail
print(vars(e))

View file

@ -664,7 +664,7 @@ async def test_key_crossing_budget():
response = await chat_completion(session=session, key=key)
pytest.fail("Should have failed - Key crossed it's budget")
except Exception as e:
assert "ExceededTokenBudget: Current spend for token:" in str(e)
assert "Budget has been exceeded!" in str(e)
@pytest.mark.skip(reason="AWS Suspended Account")

View file

@ -578,4 +578,4 @@ async def test_users_in_team_budget():
except Exception as e:
print("got exception, this is expected")
print(e)
assert "Crossed spend within team" in str(e)
assert "Budget has been exceeded" in str(e)