mirror of
https://github.com/tiennm99/litellm.git
synced 2026-07-12 21:04:10 +00:00
yuneng-jiang
7d1bd9d9f4
* fix(reset_budget): write only {spend, budget_reset_at} and stop pre-zeroing counter
ResetBudgetJob's batched update_data path shipped the full key/user/team
model on each reset. Prisma rejects object_permission_id and budget_limits
on the update input type, so any row carrying those fields detonated the
entire batch -- spend never reset, budget_reset_at never advanced. After
v1.84.0 started populating object_permission_id on UI-created keys, this
fires routinely.
_reset_budget_common also zeroed the cross-pod spend counter before the
DB write, so failed resets left enforcement reading 0 from the counter
while the DB still held the over-budget spend, admitting requests past
the cap until the counter naturally re-saturated from new reservations.
Switch the write to per-row narrow updates ({spend, budget_reset_at})
via db.batch_, and move the counter invalidation out of
_reset_budget_common so it only fires after the DB write commits. On
DB-write failure the counter is left untouched, enforcement continues
to block, and the next scheduler tick can retry without leaving a
bypass window.
Fixes #27730.
* fix(reset_budget): address Greptile review on #29358
- Strengthen the bypass-half regression test: replace the for-loop over
call_args_list (vacuously true when empty) with assert_not_called(),
so the test would actually flag a re-introduction of counter-zeroing
via any code path.
- Add the same explanatory docstring on _write_user_reset_updates and
_write_team_reset_updates that _write_key_reset_updates already has,
so all three helpers point future maintainers at #27730.
* test(reset_budget): update test_proxy_budget_reset for new batch-write path
Same shape as the previous test_reset_budget_job.py update: keys/users/teams
now write through prisma.db.batch_().<table>.update, not update_data, so the
tests need a batcher mock and updated assertions. Adds:
- _wire_batcher_for_test helper that returns a list which accumulates per-row
batch updates captured from prisma_client.db.batch_().
- _attrify helper that wraps dict fixtures so getattr(item, "token") works
alongside the dict item-access the fake_reset_* mocks rely on. The new
narrow-write helpers use getattr to pull out the row's id, and would
silently skip plain dicts otherwise.
- Updates 3 partial_failure tests to assert against the batch-call list
(rows by id, payload contains only {spend, budget_reset_at}) instead of
update_data.assert_awaited_once + data_list inspection.
- Updates test_reset_budget_continues_other_categories_on_failure: only
budget + enduser still flow through update_data; key/user/team go through
the batch path now.
- Wires the batcher mock into 3 service_logger_*_success tests so commit()
is actually awaitable and the success hook fires.
These tests were silently passing locally only because the editable install
in .venv pointed at the main repo, not the worktree — running pytest with
PYTHONPATH overridden to the worktree (matching CI) reproduces the failures.
1212 lines
46 KiB
Python
1212 lines
46 KiB
Python
import asyncio
|
|
import os
|
|
import sys
|
|
from datetime import datetime, timedelta, timezone
|
|
from unittest.mock import AsyncMock, MagicMock, patch
|
|
|
|
import pytest
|
|
from dotenv import load_dotenv
|
|
|
|
load_dotenv()
|
|
import os
|
|
|
|
from litellm.proxy._types import LiteLLM_BudgetTableFull
|
|
|
|
sys.path.insert(
|
|
0, os.path.abspath("../..")
|
|
) # Adds the parent directory to the system path
|
|
|
|
from litellm.proxy.common_utils.reset_budget_job import ResetBudgetJob
|
|
|
|
# Note: In our "fake" items we use dicts with fields that our fake reset functions modify.
|
|
# In a real-world scenario, these would be instances of LiteLLM_VerificationToken, LiteLLM_UserTable, etc.
|
|
|
|
|
|
def _attrify(d: dict):
|
|
"""
|
|
Wrap a dict so that attribute access (`.token`, `.user_id`, `.team_id`,
|
|
etc.) works alongside the existing item-access the fake_reset_* helpers
|
|
rely on. The reset job's narrow-write helpers use `getattr(item, "token",
|
|
None)` (et al), which returns None for plain dicts — that would silently
|
|
skip the row.
|
|
"""
|
|
class _AttrDict(dict):
|
|
def __getattr__(self, k):
|
|
try:
|
|
return self[k]
|
|
except KeyError:
|
|
raise AttributeError(k)
|
|
|
|
def __setattr__(self, k, v):
|
|
self[k] = v
|
|
|
|
return _AttrDict(d)
|
|
|
|
|
|
def _wire_batcher_for_test(prisma_client):
|
|
"""
|
|
Wire prisma_client.db.batch_() to return a mock batcher whose .commit() is
|
|
awaitable and whose per-table .update() calls get captured. The reset job
|
|
writes key/user/team resets via prisma.db.batch_().<table>.update — not via
|
|
prisma_client.update_data — so tests must let that batch path complete.
|
|
|
|
Returns the list that will accumulate {table, where, data} dicts from
|
|
each captured update call.
|
|
"""
|
|
batch_calls = []
|
|
|
|
def make_batcher():
|
|
class _Table:
|
|
def __init__(self, table_name):
|
|
self._table_name = table_name
|
|
|
|
def update(self, where=None, data=None):
|
|
batch_calls.append(
|
|
{"table": self._table_name, "where": where, "data": data}
|
|
)
|
|
|
|
batcher = MagicMock()
|
|
batcher.litellm_verificationtoken = _Table("key")
|
|
batcher.litellm_usertable = _Table("user")
|
|
batcher.litellm_teamtable = _Table("team")
|
|
batcher.commit = AsyncMock(return_value=None)
|
|
return batcher
|
|
|
|
prisma_client.db.batch_ = MagicMock(side_effect=make_batcher)
|
|
return batch_calls
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_reset_budget_keys_partial_failure():
|
|
"""
|
|
Test that if one key fails to reset, the failure for that key does not block processing of the other keys.
|
|
We simulate two keys where the first fails and the second succeeds.
|
|
"""
|
|
# Arrange
|
|
key1 = {
|
|
"id": "key1",
|
|
"spend": 10.0,
|
|
"budget_duration": 60,
|
|
} # Will trigger simulated failure
|
|
key2 = {"id": "key2", "spend": 15.0, "budget_duration": 60} # Should be updated
|
|
key3 = {"id": "key3", "spend": 20.0, "budget_duration": 60} # Should be updated
|
|
key4 = {"id": "key4", "spend": 25.0, "budget_duration": 60} # Should be updated
|
|
key5 = {"id": "key5", "spend": 30.0, "budget_duration": 60} # Should be updated
|
|
key6 = {"id": "key6", "spend": 35.0, "budget_duration": 60} # Should be updated
|
|
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(
|
|
return_value=[key1, key2, key3, key4, key5, key6]
|
|
)
|
|
prisma_client.update_data = AsyncMock()
|
|
# Reset job writes key resets via prisma.db.batch_().<table>.update — not
|
|
# via update_data — so wire that path.
|
|
batch_calls = _wire_batcher_for_test(prisma_client)
|
|
|
|
# Using a dummy logging object with async hooks mocked out.
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
now = datetime.utcnow()
|
|
|
|
# token is needed because the new write path uses where={"token": ...}
|
|
# and _AttrDict makes getattr work alongside item access used by fake_reset_key.
|
|
for k in [key1, key2, key3, key4, key5, key6]:
|
|
k.setdefault("token", k["id"])
|
|
key1, key2, key3, key4, key5, key6 = (
|
|
_attrify(k) for k in [key1, key2, key3, key4, key5, key6]
|
|
)
|
|
prisma_client.get_data = AsyncMock(return_value=[key1, key2, key3, key4, key5, key6])
|
|
|
|
async def fake_reset_key(key, current_time):
|
|
if key["id"] == "key1":
|
|
# Simulate a failure on key1 (for example, this might be due to an invariant check)
|
|
raise Exception("Simulated failure for key1")
|
|
else:
|
|
# Simulate successful reset modification
|
|
key["spend"] = 0.0
|
|
# Compute a new reset time based on the budget duration
|
|
key["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=key["budget_duration"])
|
|
).isoformat()
|
|
return key
|
|
|
|
with patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_key", side_effect=fake_reset_key
|
|
) as mock_reset_key:
|
|
# Call the method; even though one key fails, the loop should process both
|
|
await job.reset_budget_for_litellm_keys()
|
|
# Allow any created tasks (logging hooks) to schedule
|
|
await asyncio.sleep(0.1)
|
|
|
|
# Assert that the helper was called for 6 keys
|
|
assert mock_reset_key.call_count == 6
|
|
|
|
# Assert that the new narrow write path got 5 batched updates (key1 failed).
|
|
# update_data must NOT have been called for keys.
|
|
prisma_client.update_data.assert_not_awaited()
|
|
key_writes = [c for c in batch_calls if c["table"] == "key"]
|
|
assert len(key_writes) == 5
|
|
written_ids = [c["where"]["token"] for c in key_writes]
|
|
assert written_ids == ["key2", "key3", "key4", "key5", "key6"]
|
|
# And every write must carry only {spend, budget_reset_at} — never the full row.
|
|
for c in key_writes:
|
|
assert set(c["data"].keys()) == {"spend", "budget_reset_at"}
|
|
assert c["data"]["spend"] == 0
|
|
|
|
# Verify that the failure logging hook was scheduled (due to the failure for key1)
|
|
failure_hook_calls = (
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args_list
|
|
)
|
|
# There should be one failure hook call for keys (with call_type "reset_budget_keys")
|
|
assert any(
|
|
call.kwargs.get("call_type") == "reset_budget_keys"
|
|
for call in failure_hook_calls
|
|
)
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_reset_budget_users_partial_failure():
|
|
"""
|
|
Test that if one user fails to reset, the reset loop still processes the other users.
|
|
We simulate two users where the first fails and the second is updated.
|
|
"""
|
|
user1 = {
|
|
"id": "user1",
|
|
"spend": 20.0,
|
|
"budget_duration": 120,
|
|
} # Will trigger simulated failure
|
|
user2 = {"id": "user2", "spend": 25.0, "budget_duration": 120} # Should be updated
|
|
user3 = {"id": "user3", "spend": 30.0, "budget_duration": 120} # Should be updated
|
|
user4 = {"id": "user4", "spend": 35.0, "budget_duration": 120} # Should be updated
|
|
user5 = {"id": "user5", "spend": 40.0, "budget_duration": 120} # Should be updated
|
|
user6 = {"id": "user6", "spend": 45.0, "budget_duration": 120} # Should be updated
|
|
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(
|
|
return_value=[user1, user2, user3, user4, user5, user6]
|
|
)
|
|
prisma_client.update_data = AsyncMock()
|
|
batch_calls = _wire_batcher_for_test(prisma_client)
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
# user_id required for the new write path's where clause; _AttrDict so
|
|
# getattr(u, 'user_id') works alongside the dict access fake_reset_user uses.
|
|
for u in [user1, user2, user3, user4, user5, user6]:
|
|
u.setdefault("user_id", u["id"])
|
|
user1, user2, user3, user4, user5, user6 = (
|
|
_attrify(u) for u in [user1, user2, user3, user4, user5, user6]
|
|
)
|
|
prisma_client.get_data = AsyncMock(return_value=[user1, user2, user3, user4, user5, user6])
|
|
|
|
async def fake_reset_user(user, current_time):
|
|
if user["id"] == "user1":
|
|
raise Exception("Simulated failure for user1")
|
|
else:
|
|
user["spend"] = 0.0
|
|
user["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=user["budget_duration"])
|
|
).isoformat()
|
|
return user
|
|
|
|
with patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_user", side_effect=fake_reset_user
|
|
) as mock_reset_user:
|
|
await job.reset_budget_for_litellm_users()
|
|
await asyncio.sleep(0.1)
|
|
|
|
assert mock_reset_user.call_count == 6
|
|
prisma_client.update_data.assert_not_awaited()
|
|
user_writes = [c for c in batch_calls if c["table"] == "user"]
|
|
assert len(user_writes) == 5
|
|
written_ids = [c["where"]["user_id"] for c in user_writes]
|
|
assert written_ids == ["user2", "user3", "user4", "user5", "user6"]
|
|
for c in user_writes:
|
|
assert set(c["data"].keys()) == {"spend", "budget_reset_at"}
|
|
assert c["data"]["spend"] == 0
|
|
|
|
failure_hook_calls = (
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args_list
|
|
)
|
|
assert any(
|
|
call.kwargs.get("call_type") == "reset_budget_users"
|
|
for call in failure_hook_calls
|
|
)
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_reset_budget_endusers_partial_failure():
|
|
"""
|
|
Test that if one enduser fails to reset, the reset loop still processes the other endusers.
|
|
We simulate six endsers where the first fails and the others are updated.
|
|
"""
|
|
user1 = {
|
|
"user_id": "user1",
|
|
"spend": 20.0,
|
|
"budget_id": "budget1",
|
|
} # Will trigger simulated failure
|
|
user2 = {
|
|
"user_id": "user2",
|
|
"spend": 25.0,
|
|
"budget_id": "budget1",
|
|
} # Should be updated
|
|
user3 = {
|
|
"user_id": "user3",
|
|
"spend": 30.0,
|
|
"budget_id": "budget1",
|
|
} # Should be updated
|
|
user4 = {
|
|
"user_id": "user4",
|
|
"spend": 35.0,
|
|
"budget_id": "budget1",
|
|
} # Should be updated
|
|
user5 = {
|
|
"user_id": "user5",
|
|
"spend": 40.0,
|
|
"budget_id": "budget1",
|
|
} # Should be updated
|
|
user6 = {
|
|
"user_id": "user6",
|
|
"spend": 45.0,
|
|
"budget_id": "budget1",
|
|
} # Should be updated
|
|
|
|
budget1 = LiteLLM_BudgetTableFull(
|
|
**{
|
|
"budget_id": "budget1",
|
|
"max_budget": 65.0,
|
|
"budget_duration": "2d",
|
|
"created_at": datetime.now(timezone.utc) - timedelta(days=3),
|
|
}
|
|
)
|
|
|
|
prisma_client = MagicMock()
|
|
|
|
async def get_data_mock(table_name, *args, **kwargs):
|
|
if table_name == "budget":
|
|
return [budget1]
|
|
elif table_name == "enduser":
|
|
return [user1, user2, user3, user4, user5, user6]
|
|
return []
|
|
|
|
prisma_client.get_data = AsyncMock()
|
|
prisma_client.get_data.side_effect = get_data_mock
|
|
|
|
prisma_client.update_data = AsyncMock()
|
|
# Mock db.litellm_verificationtoken.update_many (used by reset_budget_for_keys_linked_to_budgets)
|
|
prisma_client.db.litellm_verificationtoken.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_organizationtable.update_many (used by reset_budget_for_orgs_linked_to_budgets)
|
|
prisma_client.db.litellm_organizationtable.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_tagtable.update_many (used by reset_budget_for_tags_linked_to_budgets)
|
|
prisma_client.db.litellm_tagtable.update_many = AsyncMock(return_value={"count": 0})
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_enduser(enduser):
|
|
if enduser["user_id"] == "user1":
|
|
raise Exception("Simulated failure for user1")
|
|
enduser["spend"] = 0.0
|
|
return enduser
|
|
|
|
async def fake_reset_team_members(budgets_to_reset):
|
|
return 1
|
|
|
|
with (
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_enduser",
|
|
side_effect=fake_reset_enduser,
|
|
) as mock_reset_enduser,
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"reset_budget_for_litellm_team_members",
|
|
side_effect=fake_reset_team_members,
|
|
) as mock_reset_team_members,
|
|
):
|
|
await job.reset_budget_for_litellm_budget_table()
|
|
await asyncio.sleep(0.1)
|
|
|
|
assert mock_reset_enduser.call_count == 6
|
|
assert prisma_client.update_data.await_count == 2
|
|
update_call = prisma_client.update_data.call_args
|
|
assert update_call.kwargs.get("table_name") == "enduser"
|
|
updated_users = update_call.kwargs.get("data_list", [])
|
|
assert len(updated_users) == 5
|
|
assert updated_users[0]["user_id"] == "user2"
|
|
assert updated_users[1]["user_id"] == "user3"
|
|
assert updated_users[2]["user_id"] == "user4"
|
|
assert updated_users[3]["user_id"] == "user5"
|
|
assert updated_users[4]["user_id"] == "user6"
|
|
|
|
failure_hook_calls = (
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args_list
|
|
)
|
|
assert any(
|
|
call.kwargs.get("call_type") == "reset_budget_endusers"
|
|
for call in failure_hook_calls
|
|
)
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_reset_budget_teams_partial_failure():
|
|
"""
|
|
Test that if one team fails to reset, the loop processes both teams and only updates the ones that succeeded.
|
|
We simulate two teams where the first fails and the second is updated.
|
|
"""
|
|
team1 = {
|
|
"id": "team1",
|
|
"spend": 30.0,
|
|
"budget_duration": 180,
|
|
} # Will trigger simulated failure
|
|
team2 = {"id": "team2", "spend": 35.0, "budget_duration": 180} # Should be updated
|
|
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=[team1, team2])
|
|
prisma_client.update_data = AsyncMock()
|
|
batch_calls = _wire_batcher_for_test(prisma_client)
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
# team_id required for the new write path's where clause; _AttrDict for getattr.
|
|
for t in [team1, team2]:
|
|
t.setdefault("team_id", t["id"])
|
|
team1, team2 = _attrify(team1), _attrify(team2)
|
|
prisma_client.get_data = AsyncMock(return_value=[team1, team2])
|
|
|
|
async def fake_reset_team(team, current_time):
|
|
if team["id"] == "team1":
|
|
raise Exception("Simulated failure for team1")
|
|
else:
|
|
team["spend"] = 0.0
|
|
team["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=team["budget_duration"])
|
|
).isoformat()
|
|
return team
|
|
|
|
with patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_team", side_effect=fake_reset_team
|
|
) as mock_reset_team:
|
|
await job.reset_budget_for_litellm_teams()
|
|
await asyncio.sleep(0.1)
|
|
|
|
assert mock_reset_team.call_count == 2
|
|
prisma_client.update_data.assert_not_awaited()
|
|
team_writes = [c for c in batch_calls if c["table"] == "team"]
|
|
assert len(team_writes) == 1
|
|
assert team_writes[0]["where"] == {"team_id": "team2"}
|
|
assert set(team_writes[0]["data"].keys()) == {"spend", "budget_reset_at"}
|
|
assert team_writes[0]["data"]["spend"] == 0
|
|
|
|
failure_hook_calls = (
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args_list
|
|
)
|
|
assert any(
|
|
call.kwargs.get("call_type") == "reset_budget_teams"
|
|
for call in failure_hook_calls
|
|
)
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_reset_budget_continues_other_categories_on_failure():
|
|
"""
|
|
Test that executing the overall reset_budget() method continues to process keys, users, and teams,
|
|
even if one of the sub-categories (here, users) experiences a partial failure.
|
|
|
|
In this simulation:
|
|
- All keys are processed successfully.
|
|
- One of the two users fails.
|
|
- All teams are processed successfully.
|
|
|
|
We then assert that:
|
|
- update_data is called for each category with the correctly updated items.
|
|
- Each get_data call is made (indicating that one failing category did not abort the others).
|
|
"""
|
|
# Arrange dummy items for each table
|
|
key1 = {"id": "key1", "spend": 10.0, "budget_duration": 60}
|
|
key2 = {"id": "key2", "spend": 15.0, "budget_duration": 60}
|
|
user1 = {
|
|
"id": "user1",
|
|
"spend": 20.0,
|
|
"budget_duration": 120,
|
|
} # Will fail in user reset
|
|
user2 = {"id": "user2", "spend": 25.0, "budget_duration": 120} # Succeeds
|
|
team1 = {"id": "team1", "spend": 30.0, "budget_duration": 180}
|
|
team2 = {"id": "team2", "spend": 35.0, "budget_duration": 180}
|
|
enduser1 = {"user_id": "user1", "spend": 25.0, "budget_id": "budget1"}
|
|
budget1 = LiteLLM_BudgetTableFull(
|
|
**{
|
|
"budget_id": "budget1",
|
|
"max_budget": 65.0,
|
|
"budget_duration": "2d",
|
|
"created_at": datetime.now(timezone.utc) - timedelta(days=3),
|
|
}
|
|
)
|
|
|
|
prisma_client = MagicMock()
|
|
|
|
async def fake_get_data(*, table_name, query_type, **kwargs):
|
|
if table_name == "key":
|
|
return [key1, key2]
|
|
elif table_name == "user":
|
|
return [user1, user2]
|
|
elif table_name == "team":
|
|
return [team1, team2]
|
|
elif table_name == "budget":
|
|
return [budget1]
|
|
elif table_name == "enduser":
|
|
return [enduser1]
|
|
return []
|
|
|
|
prisma_client.get_data = AsyncMock(side_effect=fake_get_data)
|
|
prisma_client.update_data = AsyncMock()
|
|
batch_calls = _wire_batcher_for_test(prisma_client)
|
|
# ID fields required by the new write path's where clauses; _AttrDict
|
|
# lets getattr() see them alongside the item-access fake_reset_* helpers use.
|
|
for k in [key1, key2]:
|
|
k.setdefault("token", k["id"])
|
|
for u in [user1, user2]:
|
|
u.setdefault("user_id", u["id"])
|
|
for t in [team1, team2]:
|
|
t.setdefault("team_id", t["id"])
|
|
key1, key2 = _attrify(key1), _attrify(key2)
|
|
user1, user2 = _attrify(user1), _attrify(user2)
|
|
team1, team2 = _attrify(team1), _attrify(team2)
|
|
# Mock db.litellm_verificationtoken.update_many (used by reset_budget_for_keys_linked_to_budgets)
|
|
prisma_client.db.litellm_verificationtoken.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_organizationtable.update_many (used by reset_budget_for_orgs_linked_to_budgets)
|
|
prisma_client.db.litellm_organizationtable.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_tagtable.update_many (used by reset_budget_for_tags_linked_to_budgets)
|
|
prisma_client.db.litellm_tagtable.update_many = AsyncMock(return_value={"count": 0})
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_key(key, current_time):
|
|
key["spend"] = 0.0
|
|
key["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=key["budget_duration"])
|
|
).isoformat()
|
|
return key
|
|
|
|
async def fake_reset_user(user, current_time):
|
|
if user["id"] == "user1":
|
|
raise Exception("Simulated failure for user1")
|
|
user["spend"] = 0.0
|
|
user["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=user["budget_duration"])
|
|
).isoformat()
|
|
return user
|
|
|
|
async def fake_reset_team(team, current_time):
|
|
team["spend"] = 0.0
|
|
team["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=team["budget_duration"])
|
|
).isoformat()
|
|
return team
|
|
|
|
async def fake_reset_enduser(enduser):
|
|
enduser["spend"] = 0.0
|
|
return enduser
|
|
|
|
async def fake_reset_team_members(budgets_to_reset):
|
|
return 1
|
|
|
|
with (
|
|
patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_key", side_effect=fake_reset_key
|
|
) as mock_reset_key,
|
|
patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_user", side_effect=fake_reset_user
|
|
) as mock_reset_user,
|
|
patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_team", side_effect=fake_reset_team
|
|
) as mock_reset_team,
|
|
patch.object(
|
|
ResetBudgetJob, "_reset_budget_for_enduser", side_effect=fake_reset_enduser
|
|
) as mock_reset_enduser,
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"reset_budget_for_litellm_team_members",
|
|
side_effect=fake_reset_team_members,
|
|
) as mock_reset_team_members,
|
|
):
|
|
# Call the overall reset_budget method.
|
|
await job.reset_budget()
|
|
await asyncio.sleep(0.1)
|
|
|
|
# Verify that get_data was called for each table. We can check the table names across calls.
|
|
called_tables = {
|
|
call.kwargs.get("table_name") for call in prisma_client.get_data.await_args_list
|
|
}
|
|
if mock_reset_team_members.call_count > 0:
|
|
called_tables.add("team_membership")
|
|
assert called_tables == {
|
|
"key",
|
|
"user",
|
|
"team",
|
|
"budget",
|
|
"enduser",
|
|
"team_membership",
|
|
}
|
|
|
|
# After the fix, keys/users/teams write via prisma.db.batch_().<table>.update,
|
|
# so only budget + enduser still go through update_data.
|
|
calls = prisma_client.update_data.await_args_list
|
|
update_data_tables = [c.kwargs.get("table_name") for c in calls]
|
|
assert sorted(update_data_tables) == ["budget", "enduser"]
|
|
|
|
# Check enduser update: enduser succeed.
|
|
enduser_call = next(c for c in calls if c.kwargs.get("table_name") == "enduser")
|
|
assert len(enduser_call.kwargs.get("data_list", [])) == 1
|
|
|
|
# Check the new batch write path: 2 keys + 1 user (user1 failed) + 2 teams.
|
|
key_writes = [c for c in batch_calls if c["table"] == "key"]
|
|
user_writes = [c for c in batch_calls if c["table"] == "user"]
|
|
team_writes = [c for c in batch_calls if c["table"] == "team"]
|
|
assert len(key_writes) == 2
|
|
assert len(user_writes) == 1
|
|
assert user_writes[0]["where"] == {"user_id": "user2"}
|
|
assert len(team_writes) == 2
|
|
# Every batched write must carry only the two reset fields, never the full row.
|
|
for c in key_writes + user_writes + team_writes:
|
|
assert set(c["data"].keys()) == {"spend", "budget_reset_at"}
|
|
assert c["data"]["spend"] == 0
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Additional tests for service logger behavior (keys, users, teams, endusers)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_keys_success():
|
|
"""
|
|
Test that when resetting keys succeeds (all keys are updated) the service
|
|
logger success hook is called with the correct event metadata and no exception is logged.
|
|
"""
|
|
keys = [
|
|
{"id": "key1", "spend": 10.0, "budget_duration": 60, "token": "key1"},
|
|
{"id": "key2", "spend": 15.0, "budget_duration": 60, "token": "key2"},
|
|
]
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=keys)
|
|
prisma_client.update_data = AsyncMock()
|
|
_wire_batcher_for_test(prisma_client)
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_key(key, current_time):
|
|
key["spend"] = 0.0
|
|
key["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=key["budget_duration"])
|
|
).isoformat()
|
|
return key
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_key",
|
|
side_effect=fake_reset_key,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_keys()
|
|
# Allow async logging task to complete
|
|
await asyncio.sleep(0.1)
|
|
mock_verbose_exc.assert_not_called()
|
|
|
|
# Verify success hook call
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_success_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_keys_found") == len(keys)
|
|
assert event_metadata.get("num_keys_updated") == len(keys)
|
|
assert event_metadata.get("num_keys_failed") == 0
|
|
# Failure hook should not be executed.
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_keys_failure():
|
|
"""
|
|
Test that when a key reset fails the service logger failure hook is called,
|
|
the event metadata reflects the number of keys processed, and that the verbose
|
|
logger exception is called.
|
|
"""
|
|
keys = [
|
|
{"id": "key1", "spend": 10.0, "budget_duration": 60},
|
|
{"id": "key2", "spend": 15.0, "budget_duration": 60},
|
|
]
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=keys)
|
|
prisma_client.update_data = AsyncMock()
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_key(key, current_time):
|
|
if key["id"] == "key1":
|
|
raise Exception("Simulated failure for key1")
|
|
key["spend"] = 0.0
|
|
key["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=key["budget_duration"])
|
|
).isoformat()
|
|
return key
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_key",
|
|
side_effect=fake_reset_key,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_keys()
|
|
await asyncio.sleep(0.1)
|
|
# Expect at least one exception logged (the inner error and the outer catch)
|
|
assert mock_verbose_exc.call_count >= 1
|
|
# Verify exception was logged with correct message
|
|
assert any(
|
|
"Failed to reset budget for key" in str(call.args)
|
|
for call in mock_verbose_exc.call_args_list
|
|
)
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_keys_found") == len(keys)
|
|
keys_found_str = event_metadata.get("keys_found", "")
|
|
assert "key1" in keys_found_str
|
|
# Success hook should not be called.
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_users_success():
|
|
"""
|
|
Test that when resetting users succeeds the service logger success hook is called with
|
|
the correct metadata and no exception is logged.
|
|
"""
|
|
users = [
|
|
{"id": "user1", "spend": 20.0, "budget_duration": 120, "user_id": "user1"},
|
|
{"id": "user2", "spend": 25.0, "budget_duration": 120, "user_id": "user2"},
|
|
]
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=users)
|
|
prisma_client.update_data = AsyncMock()
|
|
_wire_batcher_for_test(prisma_client)
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_user(user, current_time):
|
|
user["spend"] = 0.0
|
|
user["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=user["budget_duration"])
|
|
).isoformat()
|
|
return user
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_user",
|
|
side_effect=fake_reset_user,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_users()
|
|
await asyncio.sleep(0.1)
|
|
mock_verbose_exc.assert_not_called()
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_success_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_users_found") == len(users)
|
|
assert event_metadata.get("num_users_updated") == len(users)
|
|
assert event_metadata.get("num_users_failed") == 0
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_users_failure():
|
|
"""
|
|
Test that a failure during user reset calls the failure hook with appropriate metadata,
|
|
logs the exception, and does not call the success hook.
|
|
"""
|
|
users = [
|
|
{"id": "user1", "spend": 20.0, "budget_duration": 120},
|
|
{"id": "user2", "spend": 25.0, "budget_duration": 120},
|
|
]
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=users)
|
|
prisma_client.update_data = AsyncMock()
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_user(user, current_time):
|
|
if user["id"] == "user1":
|
|
raise Exception("Simulated failure for user1")
|
|
user["spend"] = 0.0
|
|
user["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=user["budget_duration"])
|
|
).isoformat()
|
|
return user
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_user",
|
|
side_effect=fake_reset_user,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_users()
|
|
await asyncio.sleep(0.1)
|
|
# Verify exception logging
|
|
assert mock_verbose_exc.call_count >= 1
|
|
# Verify exception was logged with correct message
|
|
assert any(
|
|
"Failed to reset budget for user" in str(call.args)
|
|
for call in mock_verbose_exc.call_args_list
|
|
)
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_users_found") == len(users)
|
|
users_found_str = event_metadata.get("users_found", "")
|
|
assert "user1" in users_found_str
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_teams_success():
|
|
"""
|
|
Test that when resetting teams is successful the service logger success hook is called with
|
|
the proper metadata and nothing is logged as an exception.
|
|
"""
|
|
teams = [
|
|
{"id": "team1", "spend": 30.0, "budget_duration": 180, "team_id": "team1"},
|
|
{"id": "team2", "spend": 35.0, "budget_duration": 180, "team_id": "team2"},
|
|
]
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=teams)
|
|
prisma_client.update_data = AsyncMock()
|
|
_wire_batcher_for_test(prisma_client)
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_team(team, current_time):
|
|
team["spend"] = 0.0
|
|
team["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=team["budget_duration"])
|
|
).isoformat()
|
|
return team
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_team",
|
|
side_effect=fake_reset_team,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_teams()
|
|
await asyncio.sleep(0.1)
|
|
mock_verbose_exc.assert_not_called()
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_success_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_teams_found") == len(teams)
|
|
assert event_metadata.get("num_teams_updated") == len(teams)
|
|
assert event_metadata.get("num_teams_failed") == 0
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_teams_failure():
|
|
"""
|
|
Test that a failure during team reset triggers the failure hook with proper metadata,
|
|
results in an exception log and no success hook call.
|
|
"""
|
|
teams = [
|
|
{"id": "team1", "spend": 30.0, "budget_duration": 180},
|
|
{"id": "team2", "spend": 35.0, "budget_duration": 180},
|
|
]
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(return_value=teams)
|
|
prisma_client.update_data = AsyncMock()
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_team(team, current_time):
|
|
if team["id"] == "team1":
|
|
raise Exception("Simulated failure for team1")
|
|
team["spend"] = 0.0
|
|
team["budget_reset_at"] = (
|
|
current_time + timedelta(seconds=team["budget_duration"])
|
|
).isoformat()
|
|
return team
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_team",
|
|
side_effect=fake_reset_team,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_teams()
|
|
await asyncio.sleep(0.1)
|
|
# Verify exception logging
|
|
assert mock_verbose_exc.call_count >= 1
|
|
# Verify exception was logged with correct message
|
|
assert any(
|
|
"Failed to reset budget for team" in str(call.args)
|
|
for call in mock_verbose_exc.call_args_list
|
|
)
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_teams_found") == len(teams)
|
|
teams_found_str = event_metadata.get("teams_found", "")
|
|
assert "team1" in teams_found_str
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_endusers_success():
|
|
"""
|
|
Test that when resetting endusers succeeds the service logger success hook is called with
|
|
the correct metadata and no exception is logged.
|
|
"""
|
|
endusers = [
|
|
{"user_id": "user1", "spend": 25.0, "budget_id": "budget1"},
|
|
{"user_id": "user2", "spend": 25.0, "budget_id": "budget1"},
|
|
]
|
|
budgets = [
|
|
LiteLLM_BudgetTableFull(
|
|
**{
|
|
"budget_id": "budget1",
|
|
"max_budget": 65.0,
|
|
"budget_duration": "2d",
|
|
"created_at": datetime.now(timezone.utc) - timedelta(days=3),
|
|
}
|
|
)
|
|
]
|
|
|
|
async def fake_get_data(*, table_name, query_type, **kwargs):
|
|
if table_name == "budget":
|
|
return budgets
|
|
elif table_name == "enduser":
|
|
return endusers
|
|
return []
|
|
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(side_effect=fake_get_data)
|
|
prisma_client.update_data = AsyncMock()
|
|
# Mock db.litellm_verificationtoken.update_many (used by reset_budget_for_keys_linked_to_budgets)
|
|
prisma_client.db.litellm_verificationtoken.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_organizationtable.update_many (used by reset_budget_for_orgs_linked_to_budgets)
|
|
prisma_client.db.litellm_organizationtable.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_tagtable.update_many (used by reset_budget_for_tags_linked_to_budgets)
|
|
prisma_client.db.litellm_tagtable.update_many = AsyncMock(return_value={"count": 0})
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_enduser(enduser):
|
|
enduser["spend"] = 0.0
|
|
return enduser
|
|
|
|
async def fake_reset_team_members(budgets_to_reset):
|
|
return 1
|
|
|
|
with (
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_enduser",
|
|
side_effect=fake_reset_enduser,
|
|
) as mock_reset_enduser,
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"reset_budget_for_litellm_team_members",
|
|
side_effect=fake_reset_team_members,
|
|
) as mock_reset_team_members,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_budget_table()
|
|
await asyncio.sleep(0.1)
|
|
mock_verbose_exc.assert_not_called()
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_success_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_budgets_found") == len(budgets)
|
|
assert event_metadata.get("num_endusers_found") == len(endusers)
|
|
assert event_metadata.get("num_endusers_updated") == len(endusers)
|
|
assert event_metadata.get("num_endusers_failed") == 0
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_service_logger_endusers_failure():
|
|
"""
|
|
Test that a failure during enduser reset calls the failure hook with appropriate metadata,
|
|
logs the exception, and does not call the success hook.
|
|
"""
|
|
endusers = [
|
|
{"user_id": "user1", "spend": 25.0, "budget_id": "budget1"},
|
|
{"user_id": "user2", "spend": 25.0, "budget_id": "budget1"},
|
|
]
|
|
budgets = [
|
|
LiteLLM_BudgetTableFull(
|
|
**{
|
|
"budget_id": "budget1",
|
|
"max_budget": 65.0,
|
|
"budget_duration": "2d",
|
|
"created_at": datetime.now(timezone.utc) - timedelta(days=3),
|
|
}
|
|
)
|
|
]
|
|
|
|
async def fake_get_data(*, table_name, query_type, **kwargs):
|
|
if table_name == "budget":
|
|
return budgets
|
|
elif table_name == "enduser":
|
|
return endusers
|
|
return []
|
|
|
|
prisma_client = MagicMock()
|
|
prisma_client.get_data = AsyncMock(side_effect=fake_get_data)
|
|
prisma_client.update_data = AsyncMock()
|
|
# Mock db.litellm_verificationtoken.update_many (used by reset_budget_for_keys_linked_to_budgets)
|
|
prisma_client.db.litellm_verificationtoken.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_organizationtable.update_many (used by reset_budget_for_orgs_linked_to_budgets)
|
|
prisma_client.db.litellm_organizationtable.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
# Mock db.litellm_tagtable.update_many (used by reset_budget_for_tags_linked_to_budgets)
|
|
prisma_client.db.litellm_tagtable.update_many = AsyncMock(return_value={"count": 0})
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_enduser(enduser):
|
|
if enduser["user_id"] == "user1":
|
|
raise Exception("Simulated failure for user1")
|
|
enduser["spend"] = 0.0
|
|
return enduser
|
|
|
|
async def fake_reset_team_members(budgets_to_reset):
|
|
return 1
|
|
|
|
with (
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_enduser",
|
|
side_effect=fake_reset_enduser,
|
|
) as mock_reset_enduser,
|
|
patch.object(
|
|
ResetBudgetJob,
|
|
"reset_budget_for_litellm_team_members",
|
|
side_effect=fake_reset_team_members,
|
|
) as mock_reset_team_members,
|
|
):
|
|
with patch(
|
|
"litellm.proxy.common_utils.reset_budget_job.verbose_proxy_logger.exception"
|
|
) as mock_verbose_exc:
|
|
await job.reset_budget_for_litellm_budget_table()
|
|
await asyncio.sleep(0.1)
|
|
# Verify exception logging
|
|
assert mock_verbose_exc.call_count >= 1
|
|
# Verify exception was logged with correct message
|
|
assert any(
|
|
"Failed to reset budget for enduser" in str(call.args)
|
|
for call in mock_verbose_exc.call_args_list
|
|
)
|
|
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook.assert_called_once()
|
|
(
|
|
args,
|
|
kwargs,
|
|
) = proxy_logging_obj.service_logging_obj.async_service_failure_hook.call_args
|
|
event_metadata = kwargs.get("event_metadata", {})
|
|
assert event_metadata.get("num_budgets_found") == len(budgets)
|
|
assert event_metadata.get("num_endusers_found") == len(endusers)
|
|
endusers_found_str = event_metadata.get("endusers_found", "")
|
|
assert "user1" in endusers_found_str
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook.assert_not_called()
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_reset_budget_for_litellm_team_members_called():
|
|
"""
|
|
Test that when reset_budget_for_litellm_budget_table is called,
|
|
team members' budgets are also reset via reset_budget_for_litellm_team_members
|
|
"""
|
|
# Arrange
|
|
budget1 = LiteLLM_BudgetTableFull(
|
|
**{
|
|
"budget_id": "budget1",
|
|
"max_budget": 100.0,
|
|
"budget_duration": "1d",
|
|
"created_at": datetime.now(timezone.utc) - timedelta(days=2),
|
|
}
|
|
)
|
|
|
|
enduser1 = {"user_id": "user1", "spend": 25.0, "budget_id": "budget1"}
|
|
|
|
prisma_client = MagicMock()
|
|
|
|
async def fake_get_data(*, table_name, query_type, **kwargs):
|
|
if table_name == "budget":
|
|
return [budget1]
|
|
elif table_name == "enduser":
|
|
return [enduser1]
|
|
return []
|
|
|
|
prisma_client.get_data = AsyncMock(side_effect=fake_get_data)
|
|
prisma_client.update_data = AsyncMock()
|
|
|
|
# Mock the db.litellm_teammembership.update_many call
|
|
prisma_client.db = MagicMock()
|
|
prisma_client.db.litellm_teammembership = MagicMock()
|
|
prisma_client.db.litellm_teammembership.update_many = AsyncMock(
|
|
return_value={"count": 2}
|
|
)
|
|
prisma_client.db.litellm_verificationtoken.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
prisma_client.db.litellm_organizationtable.update_many = AsyncMock(
|
|
return_value={"count": 0}
|
|
)
|
|
prisma_client.db.litellm_tagtable.update_many = AsyncMock(return_value={"count": 0})
|
|
|
|
proxy_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj = MagicMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock()
|
|
proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock()
|
|
|
|
job = ResetBudgetJob(proxy_logging_obj, prisma_client)
|
|
|
|
async def fake_reset_enduser(enduser):
|
|
enduser["spend"] = 0.0
|
|
return enduser
|
|
|
|
with patch.object(
|
|
ResetBudgetJob,
|
|
"_reset_budget_for_enduser",
|
|
side_effect=fake_reset_enduser,
|
|
):
|
|
# Act
|
|
await job.reset_budget_for_litellm_budget_table()
|
|
|
|
# Assert
|
|
# Verify that the team membership update was called
|
|
prisma_client.db.litellm_teammembership.update_many.assert_called_once()
|
|
|
|
# Verify the call was made with correct parameters
|
|
call_args = prisma_client.db.litellm_teammembership.update_many.call_args
|
|
assert call_args.kwargs["where"]["budget_id"]["in"] == ["budget1"]
|
|
assert call_args.kwargs["data"]["spend"] == 0
|