diff --git a/litellm/proxy/management_endpoints/management_v1/budgets.py b/litellm/proxy/management_endpoints/management_v1/budgets.py index 1970a0797ef..19a5408496a 100644 --- a/litellm/proxy/management_endpoints/management_v1/budgets.py +++ b/litellm/proxy/management_endpoints/management_v1/budgets.py @@ -3,6 +3,7 @@ from collections.abc import Mapping, Sequence from dataclasses import dataclass from datetime import datetime +from itertools import chain from types import MappingProxyType from typing import Annotated, Final @@ -76,6 +77,32 @@ _ROW_COUNTS: Final = TypeAdapter(tuple[_RowCount, ...]) SELECTED_COLUMNS: Final = ", ".join(f'"{name}"' for name in BudgetListItem.model_fields) +# Unicode 17.0 Nd zeroes, each followed by its other nine digits in UnicodeData.txt. +_DECIMAL_ZEROES: Final = "0٠۰߀०০੦૦୦௦౦೦൦෦๐໐༠၀႐០᠐᥆᧐᪀᪐᭐᮰᱀᱐꘠꣐꤀꧐꧰꩐꯰０𐒠𐴰𐵀𑁦𑃰𑄶𑇐𑋰𑑐𑓐𑙐𑛀𑛐𑛚𑜰𑣠𑥐𑯰𑱐𑵐𑶠\U00011de0𑽐𖄰𖩠𖫀𖭐𖵰𜳰𝟎𝟘𝟢𝟬𝟶𞅀𞋰𞓰𞗱𞥐🯰" +_DECIMAL_DIGITS: Final = "".join( + chain.from_iterable(tuple(chr(ord(zero) + digit) for digit in range(10)) for zero in _DECIMAL_ZEROES) +) +_ASCII_DIGITS: Final = "0123456789" * len(_DECIMAL_ZEROES) + + +_BUDGET_DURATION_SECONDS: Final = f"""( + SELECT EXTRACT(EPOCH FROM ( + (CURRENT_TIMESTAMP AT TIME ZONE 'UTC') + + (parts[1] || ' ' || CASE parts[2] + WHEN 's' THEN 'seconds' WHEN 'm' THEN 'minutes' WHEN 'h' THEN 'hours' + WHEN 'd' THEN 'days' WHEN 'w' THEN 'weeks' WHEN 'mo' THEN 'months' END)::interval + - (CURRENT_TIMESTAMP AT TIME ZONE 'UTC') + )) FROM regexp_match( + CASE lower(trim(budget_duration)) + WHEN 'hourly' THEN '1h' WHEN 'daily' THEN '24h' + WHEN 'weekly' THEN '7d' WHEN 'monthly' THEN '30d' + ELSE CASE WHEN octet_length(budget_duration) = length(budget_duration) THEN budget_duration + ELSE translate(budget_duration, '{_DECIMAL_DIGITS}', '{_ASCII_DIGITS}') END END, + '^([0-9]+)(mo|[smhdw])' + ) AS duration(parts) +)""" + + @dataclass(frozen=True, slots=True) class PrismaBudgetListExecutor: """The database half of the budgets list. Every caller-supplied value is bound to a @@ -91,10 +118,16 @@ class PrismaBudgetListExecutor: async def find_many(self, plan: QueryPlan) -> Sequence[BudgetListItem]: clauses, params = where_sql(plan.where) + ordering: Final = ", ".join( + order_by_sql((key,)).replace("NULLS LAST", "NULLS FIRST") + if key.descending and key.field in ("max_budget", "tpm_limit", "rpm_limit", "tpd_limit") + else order_by_sql((key,)) + for key in plan.order + ).replace('"budget_duration"', _BUDGET_DURATION_SECONDS) sql: Final = ( f"SELECT {SELECTED_COLUMNS} FROM {BUDGET_TABLE}" + (f" WHERE {clauses}" if clauses else "") - + f" ORDER BY {order_by_sql(plan.order)}" + + f" ORDER BY {ordering}" + f" LIMIT ${len(params) + 1} OFFSET ${len(params) + 2}" ) return _BUDGET_ROWS.validate_python(await self.prisma_client.db.query_raw(sql, *params, plan.take, plan.skip)) @@ -111,8 +144,6 @@ def _scope(caller: UserAPIKeyAuth) -> Scope: return ScopeDenied(reason=f"Only proxy admins can list budgets, your role={caller.user_role}") -# budget_duration is deliberately absent from `sortable`: the column holds strings -# like "7d" and "30d", so a lexicographic ORDER BY puts "30d" ahead of "7d". BUDGET_FILTERS: Final[Mapping[str, FilterSpec]] = MappingProxyType( { "budget_duration": FilterSpec(type=str, ops=frozenset(("in", "is_null"))), @@ -123,7 +154,9 @@ BUDGET_FILTERS: Final[Mapping[str, FilterSpec]] = MappingProxyType( BUDGETS_LIST_SPEC: Final[ListSpec[BudgetListItem, BudgetListItem]] = ListSpec( resource="budgets", - sortable=frozenset(("budget_id", "max_budget", "tpm_limit", "rpm_limit", "tpd_limit", "created_at")), + sortable=frozenset( + ("budget_id", "max_budget", "tpm_limit", "rpm_limit", "tpd_limit", "created_at", "budget_duration") + ), searchable=frozenset(("budget_id",)), filters=BUDGET_FILTERS, default_sort=(SortKey(field="created_at", descending=True),), @@ -154,8 +187,9 @@ async def list_budgets( way to page, sort or filter it. `sort` takes a comma-separated list of `budget_id`, `max_budget`, `tpm_limit`, - `rpm_limit`, `tpd_limit` or `created_at`, each optionally prefixed with `-` for descending, - and defaults to `-created_at`. `budget_id` is appended to every sort as the + `rpm_limit`, `tpd_limit`, `created_at` or `budget_duration`, each optionally prefixed with `-` for descending, + and defaults to `-created_at`. Reset periods sort by elapsed time, with calendar months + measured from the current UTC date and unset periods last. `budget_id` is appended to every sort as the tiebreaker. `q` is a case-insensitive substring match on `budget_id`. `page_size` defaults to 50 and is capped at 100. Filters are `filter[budget_duration][in|is_null]`, `filter[max_budget][gte|lte|is_null]` diff --git a/tests/integration/management/test_budget_list_sorting.py b/tests/integration/management/test_budget_list_sorting.py new file mode 100644 index 00000000000..aae0bd6e274 --- /dev/null +++ b/tests/integration/management/test_budget_list_sorting.py @@ -0,0 +1,94 @@ +from datetime import datetime, timezone +from itertools import chain +from typing import Final +from uuid import uuid4 + +import pytest + +from litellm.litellm_core_utils.duration_parser import duration_in_seconds_from +from tests.integration._support.client import Gateway, Scenario + + +@pytest.mark.parametrize("descending", (False, True)) +def test_budget_reset_sort_orders_durations_before_pagination( + descending: bool, + gateway: Gateway, + scenario: Scenario, +) -> None: + prefix: Final = f"sort-{uuid4().hex}" + durations: Final = ("30d", "7d", "1h", "60m", "1mo", "weekly", "monthly", "٥d", "۵d", "५d", "５d", "𝟝d", "5d", None) + ids: Final = tuple( + scenario.budget(budget_id=f"{prefix}-{index}", budget_duration=duration, max_budget=10) + for index, duration in enumerate(durations) + ) + now: Final = datetime.now(timezone.utc) + expected: Final = tuple( + budget_id + for _, budget_id in sorted( + zip(durations, ids, strict=True), + key=lambda item: ( + item[0] is None, + duration_in_seconds_from(now, item[0]) * (-1 if descending else 1) if item[0] else 0, + item[1], + ), + ) + ) + pages: Final = tuple( + gateway.request( + "GET", + "/management/v1/budgets", + params={ + "q": prefix, + "sort": "-budget_duration" if descending else "budget_duration", + "page_size": "3", + "page": str(page), + }, + ) + for page in range(1, -(-len(ids) // 3) + 1) + ) + assert all(page.status_code == 200 for page in pages), [page.text for page in pages] + assert tuple(row["budget_id"] for row in chain.from_iterable(page.json()["data"] for page in pages)) == expected + filtered: Final = gateway.request( + "GET", + "/management/v1/budgets", + params={ + "q": prefix, + "sort": "budget_duration", + "filter[budget_duration][in]": "7d,30d", + }, + ) + assert filtered.status_code == 200, filtered.text + assert [row["budget_duration"] for row in filtered.json()["data"]] == ["7d", "30d"] + + +@pytest.mark.parametrize("field", ("max_budget", "tpm_limit", "rpm_limit", "tpd_limit")) +@pytest.mark.parametrize("descending", (False, True)) +def test_budget_limits_sort_unlimited_as_infinity( + field: str, descending: bool, gateway: Gateway, scenario: Scenario +) -> None: + prefix: Final = f"sort-unlimited-{uuid4().hex}" + values: Final = (20, None, 0, 3) + ids: Final = tuple(scenario.budget(budget_id=f"{prefix}-{i}", **{field: value}) for i, value in enumerate(values)) + expected: Final = tuple( + identity + for value, identity in sorted( + zip(values, ids, strict=True), + key=lambda item: float("inf") if item[0] is None else item[0], + reverse=descending, + ) + ) + pages: Final = tuple( + gateway.request( + "GET", + "/management/v1/budgets", + params={ + "q": prefix, + "sort": ("-" if descending else "") + field, + "page_size": "2", + "page": str(page), + }, + ) + for page in (1, 2) + ) + assert all(page.status_code == 200 for page in pages), [page.text for page in pages] + assert tuple(row["budget_id"] for row in chain.from_iterable(page.json()["data"] for page in pages)) == expected diff --git a/tests/unit/proxy/management_endpoints/management_v1/test_budgets.py b/tests/unit/proxy/management_endpoints/management_v1/test_budgets.py index 2b438a9d370..4b42e6404a3 100644 --- a/tests/unit/proxy/management_endpoints/management_v1/test_budgets.py +++ b/tests/unit/proxy/management_endpoints/management_v1/test_budgets.py @@ -52,7 +52,7 @@ app.include_router(router) client = TestClient(app) BUDGETS_PATH = f"{MANAGEMENT_V1_PREFIX}/budgets" -SORTABLE = ["budget_id", "created_at", "max_budget", "rpm_limit", "tpd_limit", "tpm_limit"] +SORTABLE = ["budget_duration", "budget_id", "created_at", "max_budget", "rpm_limit", "tpd_limit", "tpm_limit"] def _row(budget_id: str, **overrides: Any) -> dict[str, Any]: @@ -172,22 +172,12 @@ def test_appends_the_tiebreaker_to_an_explicit_sort(query_raw, as_proxy_admin): _get("sort=-max_budget") - assert 'ORDER BY "max_budget" DESC NULLS LAST, "budget_id" ASC NULLS LAST' in _select_call(query_raw)[0] + assert 'ORDER BY "max_budget" DESC NULLS FIRST, "budget_id" ASC NULLS LAST' in _select_call(query_raw)[0] -def test_refuses_to_sort_on_budget_duration(query_raw, as_proxy_admin): - """The column holds "7d"/"30d", so a lexicographic ORDER BY would put "30d" - before "7d" and silently mis-order the page.""" - _serve(query_raw, []) - +def test_accepts_duration_sorting(query_raw, as_proxy_admin): response = _get("sort=budget_duration") - - assert response.status_code == 400 - assert response.headers["content-type"].startswith("application/problem+json") - body = response.json() - assert "budget_duration" in body["detail"] - assert body["allowed"] == SORTABLE - query_raw.assert_not_called() + assert response.status_code == 200 def test_the_advertised_sort_fields_are_the_ones_that_work(query_raw, as_proxy_admin): @@ -346,7 +336,7 @@ def test_q_does_not_search_any_other_column(query_raw, as_proxy_admin): def test_is_null_selects_the_unlimited_budgets(query_raw, as_proxy_admin): - """"Unlimited" is max_budget IS NULL; `max_budget = 0` would be a hard zero cap.""" + """ "Unlimited" is max_budget IS NULL; `max_budget = 0` would be a hard zero cap.""" _serve(query_raw, [_row("b-unlimited", max_budget=None)]) body = _get("filter[max_budget][is_null]=true").json() diff --git a/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTable.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTable.test.tsx index ae898645de4..b4fb7586e6d 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTable.test.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTable.test.tsx @@ -134,13 +134,16 @@ describe("BudgetTable", () => { } }); - it("should not make the reset column sortable", async () => { + it("should request duration sorting from the server", async () => { const user = userEvent.setup(); - renderWithProviders(); + const onSortingChange = vi.fn(); + renderWithProviders(); await showColumn(user, "budget_duration"); const headers = screen.getAllByRole("columnheader").map((header) => header.textContent); expect(headers).toContain("Reset"); - expect(screen.queryByTestId("sort-header-budget_duration")).not.toBeInTheDocument(); + await user.click(screen.getByTestId("sort-header-budget_duration")); + expect(onSortingChange).toHaveBeenCalled(); + expect(onSortingChange.mock.calls[0][0]([])).toEqual([{ id: "budget_duration", desc: false }]); }); it("should ask the list for a new sort when a sortable header is clicked", async () => { diff --git a/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTableColumns.tsx b/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTableColumns.tsx index fb894322208..3ec291c1158 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTableColumns.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/budgets/_components/BudgetTableColumns.tsx @@ -139,8 +139,7 @@ export const getBudgetTableColumns = ({ accessorKey: "budget_duration", filterFn: serverFilter, meta: { title: "Reset" }, - // "7d"/"30d" sort lexicographically, not chronologically, so the route does not offer it. - enableSorting: false, + sortDescFirst: false, header: ({ column }) => , size: 110, cell: ({ row }) => , diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index a71933626fc..1d36f5fd9a2 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -9280,8 +9280,9 @@ export interface paths { * way to page, sort or filter it. * * `sort` takes a comma-separated list of `budget_id`, `max_budget`, `tpm_limit`, - * `rpm_limit`, `tpd_limit` or `created_at`, each optionally prefixed with `-` for descending, - * and defaults to `-created_at`. `budget_id` is appended to every sort as the + * `rpm_limit`, `tpd_limit`, `created_at` or `budget_duration`, each optionally prefixed with `-` for descending, + * and defaults to `-created_at`. Reset periods sort by elapsed time, with calendar months + * measured from the current UTC date and unset periods last. `budget_id` is appended to every sort as the * tiebreaker. `q` is a case-insensitive substring match on `budget_id`. * `page_size` defaults to 50 and is capped at 100. Filters are * `filter[budget_duration][in|is_null]`, `filter[max_budget][gte|lte|is_null]`