fix(interactions): survive a settlement install failure at boot and stop carrying request headers

litellm/interactions/background_cost_polling.py

@@ -211,7 +211,6 @@ _CARRIED_METADATA_KEYS: Final = frozenset(
         "spend_logs_metadata",
         "requester_metadata",
         "requester_ip_address",
-        "requester_custom_headers",
         "user_agent",
         "agent_id",
         "session_id",

litellm/proxy/spend_tracking/background_interaction_settlement.py

@@ -169,7 +169,13 @@ async def configure_background_interaction_settlement(
 
 
 async def install_background_interaction_settlement(prisma_client: "PrismaClient") -> None:
-    await configure_background_interaction_settlement(
-        table=BackgroundInteractionSettlementRepository(prisma_client).table,
-        claimed_by=f"{socket.gethostname()}:{os.getpid()}",
-    )
+    try:
+        await configure_background_interaction_settlement(
+            table=BackgroundInteractionSettlementRepository(prisma_client).table,
+            claimed_by=f"{socket.gethostname()}:{os.getpid()}",
+        )
+    except Exception as e:  # noqa: BLE001  # a boot step must survive any DB error; billing then settles in-process as before
+        verbose_proxy_logger.warning(
+            "Durable background interaction settlement is off on this replica, so billing settles in-process only: %s",
+            e,
+        )

tests/test_litellm/interactions/test_background_cost_polling.py

@@ -682,6 +682,23 @@ async def test_delete_settles_once_however_many_replicas_try():
     assert second_calls == []
 
 
+def test_create_context_carries_no_request_headers():
+    logging_obj = _logging_obj(
+        litellm_params={
+            "metadata": _create_metadata(
+                requester_custom_headers={"x-api-key": "sk-customer-secret"},
+                proxy_server_request={"headers": {"x-api-key": "sk-customer-secret"}},
+            )
+        }
+    )
+
+    carried = _create_context(logging_obj, "gemini").metadata
+
+    assert carried["user_api_key_team_id"] == "team-1"
+    assert "requester_custom_headers" not in carried
+    assert "proxy_server_request" not in carried
+
+
 @pytest.mark.asyncio
 async def test_restart_resumes_only_the_rows_no_replica_claimed():
     store = InMemoryBackgroundSettlementStore()

tests/test_litellm/proxy/spend_tracking/test_background_interaction_settlement.py

@@ -17,6 +17,7 @@ from litellm.interactions.background_cost_polling import (
 from litellm.litellm_core_utils.litellm_logging import Logging as LitellmLogging
 from litellm.proxy.spend_tracking.background_interaction_settlement import (
     configure_background_interaction_settlement,
+    install_background_interaction_settlement,
     PrismaBackgroundSettlementStore,
 )
 from litellm.types.interactions import InteractionsAPIResponse
@@ -233,6 +234,19 @@ async def test_configure_installs_the_store_and_resumes_the_orphaned_rows():
         configure_background_settlement_store(previous_store)
 
 
+class _PrismaClientWithoutSettlementTable:
+    pass
+
+
+@pytest.mark.asyncio
+async def test_install_keeps_booting_when_the_settlement_table_is_unreachable():
+    previous_store = bg._STORE.store
+
+    await install_background_interaction_settlement(_PrismaClientWithoutSettlementTable())
+
+    assert bg._STORE.store is previous_store
+
+
 @dataclass(frozen=True)
 class _JsonLike:
     data: object