fix(ptu): hand the prune a plain delete filter the query builder can serialise (#37571)

* fix(ptu): hand the prune a plain delete filter the query builder can serialise

The bounded sweep built its predicate as a read-only mapping view, which the query
builder refuses to serialise, so the nightly job raised as soon as a config-declared
deployment was priced. The charges were already written by then, which is why the run
looked like it had produced its rows.

The in-memory table these tests run against accepts any mapping, so only a live run
caught it. A predicate builder now returns a plain dict and is asserted as one, and the
catch-up pass has a test covering a config-declared reservation.

* refactor(ptu): build the prune predicate in one shot

Both filter shapes are known upfront, so the bounded one is constructed
directly rather than by mutating a value already declared Final.

The catch-up test took two independent clock reads, which disagree across
UTC midnight; it now derives both the reservation start and the expected
last charged day from a single read, matching the three sibling tests.

Author: yucheng-berri

litellm/proxy/spend_tracking/ptu_flat_cost_rollup.py

@@ -724,6 +724,20 @@ async def _deliver_alert(alert: "Callable[[str], Awaitable[None]] | None", messa
         verbose_proxy_logger.error("PTU rollup: could not deliver the failed-charge alert: %s", exc)
 
 
+def _prune_filter(*, date_str: str, cutoff: datetime, chunk: "tuple[str, ...] | None") -> "Mapping[str, object]":
+    """One delete statement's predicate. An absent chunk leaves the sweep unbounded.
+
+    Returns a plain dict because the query builder serialises the mapping it is handed and
+    rejects a read-only view of one.
+    """
+    return {  # mutable-ok: prisma delete filter
+        "date": date_str,
+        "api_key": PTU_SENTINEL_API_KEY,
+        "updated_at": {"lt": cutoff},  # mutable-ok: prisma comparison filter
+        **({} if chunk is None else {"model": {"in": chunk}}),  # mutable-ok: prisma membership filter
+    }
+
+
 async def _prune_unrefreshed_sentinel_rows(
     prisma_client: "PrismaClient",
     *,
@@ -752,27 +766,15 @@ async def _prune_unrefreshed_sentinel_rows(
     that many deployments would otherwise hit every night with no handler above here.
     """
     cutoff: Final = run_started - timedelta(seconds=PTU_PRUNE_SKEW_GRACE_SECONDS)
-    unbounded: Final = {  # mutable-ok: prisma delete filter
-        "date": date_str,
-        "api_key": PTU_SENTINEL_API_KEY,
-        "updated_at": {"lt": cutoff},  # mutable-ok: prisma comparison filter
-    }
     ordered: Final = () if scanned_ids is None else tuple(sorted(scanned_ids))
-    filters: Final = (
-        (unbounded,)
+    chunks: Final = (
+        (None,)
         if scanned_ids is None
         else tuple(
-            MappingProxyType(
-                {
-                    **unbounded,
-                    "model": {  # mutable-ok: prisma membership filter
-                        "in": ordered[start : start + _PRUNE_ID_CHUNK_SIZE]
-                    },
-                }
-            )
-            for start in range(0, len(ordered), _PRUNE_ID_CHUNK_SIZE)
+            ordered[start : start + _PRUNE_ID_CHUNK_SIZE] for start in range(0, len(ordered), _PRUNE_ID_CHUNK_SIZE)
         )
     )
+    filters: Final = tuple(_prune_filter(date_str=date_str, cutoff=cutoff, chunk=chunk) for chunk in chunks)
     deletions: Final = tuple(
         [await prisma_client.db.litellm_dailyteamspend.delete_many(where=where) for where in filters]
     )

tests/test_litellm/proxy/spend_tracking/test_ptu_flat_cost_rollup.py

@@ -2045,3 +2045,40 @@ def test_the_router_lookup_returns_none_outside_a_proxy():
     finally:
         if real is not None:
             sys.modules["litellm.proxy.proxy_server"] = real
+
+
+@pytest.mark.parametrize("chunk", [None, ("dep-a", "dep-b")], ids=["unbounded", "bounded"])
+def test_the_prune_filter_is_a_plain_dict(chunk):
+    """The query builder serialises the mapping it is handed and rejects a read-only view of
+    one, which the in-memory table in these tests accepts happily. Only a live run caught it."""
+    predicate = ptu_rollup._prune_filter(date_str=DAY.isoformat(), cutoff=datetime.now(timezone.utc), chunk=chunk)
+
+    assert type(predicate) is dict
+    assert type(predicate["updated_at"]) is dict
+    if chunk is None:
+        assert "model" not in predicate
+    else:
+        assert type(predicate["model"]) is dict
+        assert predicate["model"]["in"] == chunk
+
+
+@pytest.mark.asyncio
+async def test_the_catch_up_pass_reaches_a_config_declared_deployment(monkeypatch):
+    """The catch-up shares the loader, so config deployments join it without being wired in.
+    That is what prices the elapsed days of a reservation declared before today."""
+    table = _FakeSentinelTable()
+    now = datetime.now(timezone.utc)
+    started = (now - timedelta(days=3)).strftime("%Y-%m-%dT00:00:00Z")
+    entry = _router_entry(
+        model_id="cfg-back",
+        model_info={"ptu_count": 100, "cost_per_ptu_per_hour": 0.02, "team_id": "t", "ptu_effective_from": started},
+    )
+    monkeypatch.setattr(ptu_rollup, "_running_router", lambda: _router_holding(entry))
+
+    await run_scheduled_ptu_rollup(_prisma_for([], table), pod_lock_manager=_pod_lock(acquired=True))
+
+    charged = sorted(day for (_, day, _, model) in table.rows if model == "cfg-back")
+    yesterday = (now.date() - timedelta(days=1)).isoformat()
+    assert len(charged) == 3, charged
+    assert charged[-1] == yesterday
+    assert all(row["ptu_flat_cost"] == pytest.approx(48.0) for row in table.rows.values())