Invalidate spend counters on budget reset

This commit is contained in:
Michael Riad Zaky 2026-04-29 15:41:05 -07:00
parent 9f08db91f9
commit fed5f36a3d
2 changed files with 217 additions and 28 deletions

View file

@ -52,6 +52,30 @@ class ResetBudgetJob:
### RESET MULTI-WINDOW BUDGETS ###
await self.reset_budget_windows()
@staticmethod
async def _invalidate_spend_counter(counter_key: str) -> None:
"""Zero a spend counter so a DB-row reset takes effect immediately."""
try:
from litellm.proxy.proxy_server import spend_counter_cache
spend_counter_cache.in_memory_cache.set_cache(key=counter_key, value=0.0)
if spend_counter_cache.redis_cache is not None:
try:
await spend_counter_cache.redis_cache.async_set_cache(
key=counter_key, value=0.0
)
except Exception as redis_err:
verbose_proxy_logger.warning(
"Failed to reset spend counter %s in Redis: %s. "
"Budget may be over-enforced until counter expires.",
counter_key,
redis_err,
)
except Exception as e:
verbose_proxy_logger.warning(
"Failed to reset spend counter %s: %s", counter_key, e
)
async def reset_budget_for_litellm_team_members(
self, budgets_to_reset: List[LiteLLM_BudgetTableFull]
):
@ -64,37 +88,17 @@ class ResetBudgetJob:
if budget.budget_id is not None
]
# Reset spend counters for affected team members.
# Reset Redis directly so a transient failure doesn't leave stale
# counters that get_current_spend would read as authoritative.
try:
from litellm.proxy.proxy_server import spend_counter_cache
memberships = await self.prisma_client.db.litellm_teammembership.find_many(
where={"budget_id": {"in": budget_ids}}
)
for m in memberships:
counter_key = f"spend:team_member:{m.user_id}:{m.team_id}"
# Always reset in-memory
spend_counter_cache.in_memory_cache.set_cache(
key=counter_key, value=0.0
await self._invalidate_spend_counter(
f"spend:team_member:{m.user_id}:{m.team_id}"
)
# Explicitly reset Redis with warning on failure
if spend_counter_cache.redis_cache is not None:
try:
await spend_counter_cache.redis_cache.async_set_cache(
key=counter_key, value=0.0
)
except Exception as redis_err:
verbose_proxy_logger.warning(
"Failed to reset team member spend counter in Redis %s: %s. "
"Budget may be over-enforced until counter expires.",
counter_key,
redis_err,
)
except Exception as e:
verbose_proxy_logger.warning(
"Failed to reset team member spend counters: %s", e
"Failed to fetch team memberships for counter invalidation: %s", e
)
return await self.prisma_client.db.litellm_teammembership.update_many(
@ -126,12 +130,25 @@ class ResetBudgetJob:
if not budget_ids:
return
where_clause: dict = {
"budget_id": {"in": budget_ids},
"budget_duration": None, # only keys without their own reset schedule
"spend": {"gt": 0}, # only reset keys that have accumulated spend
}
try:
keys = await self.prisma_client.db.litellm_verificationtoken.find_many(
where=where_clause
)
for k in keys:
await self._invalidate_spend_counter(f"spend:key:{k.token}")
except Exception as e:
verbose_proxy_logger.warning(
"Failed to fetch keys for counter invalidation: %s", e
)
return await self.prisma_client.db.litellm_verificationtoken.update_many(
where={
"budget_id": {"in": budget_ids},
"budget_duration": None, # only keys without their own reset schedule
"spend": {"gt": 0}, # only reset keys that have accumulated spend
},
where=where_clause,
data={
"spend": 0,
},
@ -360,6 +377,10 @@ class ResetBudgetJob:
)
if updated_keys:
for k in updated_keys:
token = getattr(k, "token", None)
if token:
await self._invalidate_spend_counter(f"spend:key:{token}")
await self.prisma_client.update_data(
query_type="update_many",
data_list=updated_keys,
@ -445,6 +466,12 @@ class ResetBudgetJob:
"Updated users %s", json.dumps(updated_users, indent=4, default=str)
)
if updated_users:
for u in updated_users:
user_id = getattr(u, "user_id", None)
if user_id:
await self._invalidate_spend_counter(
f"spend:user:{user_id}"
)
await self.prisma_client.update_data(
query_type="update_many",
data_list=updated_users,
@ -536,6 +563,12 @@ class ResetBudgetJob:
"Updated teams %s", json.dumps(updated_teams, indent=4, default=str)
)
if updated_teams:
for t in updated_teams:
team_id = getattr(t, "team_id", None)
if team_id:
await self._invalidate_spend_counter(
f"spend:team:{team_id}"
)
await self.prisma_client.update_data(
query_type="update_many",
data_list=updated_teams,

View file

@ -1049,3 +1049,159 @@ def test_reset_budget_windows_query_error_does_not_break_team_path(monkeypatch):
asyncio.run(job.reset_budget_windows()) # must not raise
prisma_client.db.litellm_teamtable.update.assert_awaited_once()
# ---------------------------------------------------------------------------
# Counter invalidation on budget reset
# ---------------------------------------------------------------------------
def _make_counter_invalidation_job(monkeypatch):
"""Stub spend_counter_cache so we can observe invalidation calls."""
spend_counter_cache = MagicMock()
spend_counter_cache.in_memory_cache.set_cache = MagicMock()
spend_counter_cache.redis_cache = MagicMock()
spend_counter_cache.redis_cache.async_set_cache = AsyncMock()
fake_module = types.ModuleType("litellm.proxy.proxy_server")
fake_module.spend_counter_cache = spend_counter_cache
monkeypatch.setitem(sys.modules, "litellm.proxy.proxy_server", fake_module)
return spend_counter_cache
def test_reset_budget_for_team_members_invalidates_redis_counter(monkeypatch):
"""Team-member budget reset clears the Redis spend counter."""
counter_cache = _make_counter_invalidation_job(monkeypatch)
expired_budget = type("B", (), {"budget_id": "budget-1"})
membership = type(
"Membership",
(),
{"user_id": "alice", "team_id": "team-x", "budget_id": "budget-1"},
)
prisma_client = MagicMock()
prisma_client.db.litellm_teammembership.find_many = AsyncMock(
return_value=[membership]
)
prisma_client.db.litellm_teammembership.update_many = AsyncMock(
return_value={"count": 1}
)
job = ResetBudgetJob(proxy_logging_obj=MagicMock(), prisma_client=prisma_client)
asyncio.run(job.reset_budget_for_litellm_team_members([expired_budget]))
counter_cache.in_memory_cache.set_cache.assert_any_call(
key="spend:team_member:alice:team-x", value=0.0
)
counter_cache.redis_cache.async_set_cache.assert_any_await(
key="spend:team_member:alice:team-x", value=0.0
)
def test_reset_budget_for_keys_invalidates_redis_counter(
reset_budget_job, mock_prisma_client, monkeypatch
):
"""Key budget reset must clear the Redis spend counter."""
counter_cache = _make_counter_invalidation_job(monkeypatch)
now = datetime.now(timezone.utc)
mock_prisma_client.data["key"] = [
type(
"Key",
(),
{
"spend": 100.0,
"budget_duration": "30d",
"budget_reset_at": now,
"id": "key-1",
"token": "sk-abc",
},
)
]
asyncio.run(reset_budget_job.reset_budget_for_litellm_keys())
counter_cache.in_memory_cache.set_cache.assert_any_call(
key="spend:key:sk-abc", value=0.0
)
def test_reset_budget_for_users_invalidates_redis_counter(
reset_budget_job, mock_prisma_client, monkeypatch
):
"""User budget reset must clear the Redis spend counter."""
counter_cache = _make_counter_invalidation_job(monkeypatch)
now = datetime.now(timezone.utc)
mock_prisma_client.data["user"] = [
type(
"User",
(),
{
"spend": 50.0,
"budget_duration": "7d",
"budget_reset_at": now,
"id": "user-1",
"user_id": "alice",
},
)
]
asyncio.run(reset_budget_job.reset_budget_for_litellm_users())
counter_cache.in_memory_cache.set_cache.assert_any_call(
key="spend:user:alice", value=0.0
)
def test_reset_budget_for_teams_invalidates_redis_counter(
reset_budget_job, mock_prisma_client, monkeypatch
):
"""Team budget reset must clear the Redis spend counter."""
counter_cache = _make_counter_invalidation_job(monkeypatch)
now = datetime.now(timezone.utc)
mock_prisma_client.data["team"] = [
type(
"Team",
(),
{
"spend": 200.0,
"budget_duration": "1mo",
"budget_reset_at": now,
"id": "team-1",
"team_id": "team-x",
},
)
]
asyncio.run(reset_budget_job.reset_budget_for_litellm_teams())
counter_cache.in_memory_cache.set_cache.assert_any_call(
key="spend:team:team-x", value=0.0
)
def test_reset_budget_for_keys_linked_to_budgets_invalidates_redis_counter(monkeypatch):
"""Resetting keys via budget tier must clear each linked key's counter."""
counter_cache = _make_counter_invalidation_job(monkeypatch)
expired_budget = type("B", (), {"budget_id": "budget-1"})
linked_key = type("Key", (), {"token": "sk-linked"})
prisma_client = MagicMock()
prisma_client.db.litellm_verificationtoken.find_many = AsyncMock(
return_value=[linked_key]
)
prisma_client.db.litellm_verificationtoken.update_many = AsyncMock(
return_value={"count": 1}
)
job = ResetBudgetJob(proxy_logging_obj=MagicMock(), prisma_client=prisma_client)
asyncio.run(job.reset_budget_for_keys_linked_to_budgets([expired_budget]))
counter_cache.in_memory_cache.set_cache.assert_any_call(
key="spend:key:sk-linked", value=0.0
)