fix(proxy_server.py): use consistent 400-status code error code for exceeded budget errors

standardizes error code for budget exceeded errors to status code 400

litellm/proxy/auth/auth_checks.py

@@ -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
 

litellm/proxy/proxy_server.py

@@ -1202,8 +1202,9 @@ 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
@@ -1239,8 +1240,9 @@ 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
@@ -1264,8 +1266,9 @@ 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

litellm/tests/test_key_generate_prisma.py

@@ -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))
 
 

tests/test_keys.py

@@ -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")