fix(interactions): drop the stored request context once a settlement row is settled

litellm/proxy/spend_tracking/background_interaction_settlement.py

@@ -1,7 +1,7 @@
 import asyncio
 import os
 import socket
-from collections.abc import Awaitable, Sequence
+from collections.abc import Awaitable, Mapping, Sequence
 from dataclasses import dataclass
 from datetime import datetime, timezone
 from typing import TYPE_CHECKING, Final, Protocol
@@ -69,6 +69,7 @@ class _Claim(TypedDict):
 class _Outcome(TypedDict):
     settled_at: ReadOnly[datetime]
     outcome: ReadOnly[SettlementOutcome]
+    create_context: ReadOnly[object]
 
 
 class _SettlementTableActions(Protocol):
@@ -85,6 +86,12 @@ def _settlement_table(prisma_client: "PrismaClient") -> _SettlementTableActions:
     return BackgroundInteractionSettlementRepository(prisma_client).table
 
 
+def _json(data: Mapping[str, object]) -> object:
+    from prisma import Json  # noqa: PLC0415  # local import: prisma may be ungenerated at module load in some tools
+
+    return Json(data)
+
+
 def _pending_rows(rows: Sequence[_SettlementRow]) -> tuple[PendingBackgroundInteraction, ...]:
     return tuple(pending for row in rows for pending in _pending_row(row))
 
@@ -114,13 +121,11 @@ class PrismaBackgroundSettlementStore:
     claimed_by: str
 
     async def register(self, pending: PendingBackgroundInteraction) -> None:
-        from prisma import Json  # noqa: PLC0415  # local import: prisma may be ungenerated at module load in some tools
-
         await self.table.create(
             data=_NewSettlementRow(
                 interaction_id=pending.interaction_id,
                 custom_llm_provider=pending.custom_llm_provider,
-                create_context=Json(pending.create_context.model_dump(mode="json")),
+                create_context=_json(pending.create_context.model_dump(mode="json")),
                 created_at=pending.created_at,
             )
         )
@@ -144,7 +149,7 @@ class PrismaBackgroundSettlementStore:
 
     async def record_outcome(self, interaction_id: str, outcome: SettlementOutcome) -> None:
         await self.table.update_many(
-            data=_Outcome(settled_at=datetime.now(timezone.utc), outcome=outcome),
+            data=_Outcome(settled_at=datetime.now(timezone.utc), outcome=outcome, create_context=_json({})),
             where=_RowKey(interaction_id=interaction_id),
         )
 

tests/test_litellm/proxy/spend_tracking/test_background_interaction_settlement.py

@@ -74,7 +74,7 @@ class _FakeSettlementTable:
         matched = self._matching(where)
         for row in matched:
             for column, value in data.items():
-                setattr(row, column, value)
+                setattr(row, column, getattr(value, "data", value) if column == "create_context" else value)
         return len(matched)
 
     def _matching(self, where) -> list:
@@ -184,11 +184,12 @@ async def test_unclaimed_skips_claimed_and_unreadable_rows():
 
 
 @pytest.mark.asyncio
-async def test_record_outcome_keeps_the_audit_trail_on_the_row():
+async def test_record_outcome_keeps_the_audit_trail_and_drops_the_stored_request_context():
     table = _FakeSettlementTable()
     store = PrismaBackgroundSettlementStore(table=table, claimed_by="replica-a:1")
     await store.register(_pending("interactions/bg-1"))
     assert await store.claim("interactions/bg-1")
+    assert table.rows["interactions/bg-1"].create_context
 
     await store.record_outcome("interactions/bg-1", "billed")
 
@@ -196,6 +197,7 @@ async def test_record_outcome_keeps_the_audit_trail_on_the_row():
     assert row.outcome == "billed"
     assert row.settled_at is not None
     assert row.claimed_at <= row.settled_at
+    assert row.create_context == {}
 
 
 @pytest.mark.asyncio