diff --git a/litellm/llms/mongodb/vector_stores/transformation.py b/litellm/llms/mongodb/vector_stores/transformation.py index 92965ab745b..606da9cffb4 100644 --- a/litellm/llms/mongodb/vector_stores/transformation.py +++ b/litellm/llms/mongodb/vector_stores/transformation.py @@ -1,4 +1,5 @@ from collections.abc import Mapping, Sequence +from ipaddress import ip_address from math import isfinite from types import MappingProxyType from typing import TYPE_CHECKING, Final, Literal, NoReturn @@ -196,7 +197,7 @@ class MongoDBVectorStoreConfig(BaseQueryEmbeddingVectorStoreConfig): def get_complete_url(self, api_base: str | None, litellm_params: dict[str, object]) -> str: if not api_base: - raise config_error("MongoDB sidecar api_base is required, for example http://mongodb-sidecar:8080.") + raise config_error("MongoDB sidecar api_base is required, for example http://127.0.0.1:8080.") try: parsed: Final = urlsplit(api_base) valid: Final = parsed.scheme in ("http", "https") and bool(parsed.hostname) and parsed.port != 0 @@ -206,6 +207,17 @@ class MongoDBVectorStoreConfig(BaseQueryEmbeddingVectorStoreConfig): raise config_error( "MongoDB sidecar api_base must be an HTTP or HTTPS URL without credentials, query, or fragment." ) + if parsed.scheme == "http": + try: + loopback: Final = ip_address(parsed.hostname or "").is_loopback + except ValueError: + raise config_error( + "MongoDB sidecar requires HTTPS. HTTP is supported only for a loopback IP such as 127.0.0.1." + ) from None + if not loopback: + raise config_error( + "MongoDB sidecar requires HTTPS. HTTP is supported only for a loopback IP such as 127.0.0.1." + ) return api_base.rstrip("/") @staticmethod @@ -215,7 +227,10 @@ class MongoDBVectorStoreConfig(BaseQueryEmbeddingVectorStoreConfig): return 30_000 if not isinstance(seconds, (int, float)) or not isfinite(seconds) or seconds <= 0: raise config_error("MongoDB search timeout must be a positive finite number.") - return max(1, min(int(seconds * 1000), 30_000)) + try: + return max(1, int(seconds * 1000)) + except (ValueError, OverflowError): + raise config_error("MongoDB search timeout must be a positive finite number.") from None @classmethod def _params( diff --git a/tests/test_litellm/llms/mongodb/vector_stores/test_mongodb_transformation.py b/tests/test_litellm/llms/mongodb/vector_stores/test_mongodb_transformation.py index bca2b544673..9de473fb1f0 100644 --- a/tests/test_litellm/llms/mongodb/vector_stores/test_mongodb_transformation.py +++ b/tests/test_litellm/llms/mongodb/vector_stores/test_mongodb_transformation.py @@ -7,10 +7,10 @@ import httpx import pytest import litellm -from litellm.llms.custom_httpx.http_handler import HTTPHandler +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler from litellm.llms.mongodb.vector_stores.transformation import MongoDBVectorStoreConfig from litellm.types.utils import EmbeddingResponse -from litellm.types.vector_stores import VectorStoreSearchOptionalRequestParams +from litellm.types.vector_stores import VectorStoreSearchOptionalRequestParams, VectorStoreSearchResponse BASE_PARAMS: Final = { "api_base": "https://sidecar.example/prefix", @@ -131,52 +131,92 @@ def test_invalid_search_is_rejected_before_embedding( (200, {**RESULT, "data": [{"score": "wrong"}]}, litellm.ServiceUnavailableError), (0, {}, litellm.Timeout), (-1, {}, litellm.BadRequestError), + (-2, {"api_base": "http://sidecar.example"}, litellm.BadRequestError), + (-2, {"api_base": "http://10.0.0.10:8080"}, litellm.BadRequestError), + (-2, {"api_base": "http://localhost:8080"}, litellm.BadRequestError), (200, RESULT, None), ], ) -def test_public_sdk_preserves_http_errors_response_and_timeout( - status: int, body: Mapping[str, object], error_type: type[Exception] | None +@pytest.mark.parametrize("asynchronous", [False, True]) +@pytest.mark.parametrize("timeout", [0.75, 120.0]) +@pytest.mark.parametrize("api_base", ["https://sidecar.example/prefix", "http://127.0.0.1:8080", "http://[::1]:8080"]) +@pytest.mark.asyncio +async def test_public_sdk_preserves_http_errors_response_and_timeout( + status: int, + body: Mapping[str, object], + error_type: type[Exception] | None, + asynchronous: bool, + timeout: float, + api_base: str, ) -> None: executor: Final = RecordingEmbeddingExecutor() if status == -1: - with pytest.raises(litellm.BadRequestError, match="search-only"): - litellm.vector_stores.create(custom_llm_provider="mongodb") - executor.call.assert_not_called() + if asynchronous: + with pytest.raises(litellm.BadRequestError, match="search-only"): + await litellm.vector_stores.acreate(custom_llm_provider="mongodb") + else: + with pytest.raises(litellm.BadRequestError, match="search-only"): + litellm.vector_stores.create(custom_llm_provider="mongodb") return - - def respond(request: httpx.Request) -> httpx.Response: - assert request.url == "https://sidecar.example/prefix/v1/vector_stores/policy_index/search" - assert request.headers["authorization"] == "Bearer test-sidecar-key" - assert request.extensions["timeout"]["read"] == 0.75 - payload: Final = json.loads(request.content) - assert payload["timeout_ms"] == 750 - assert payload["query_vector"] == [0.1, 0.2, 0.3] - if status == 0: - raise httpx.ReadTimeout("timed out", request=request) - return httpx.Response(status, json=body) - - with httpx.Client(transport=httpx.MockTransport(respond)) as transport: - client: Final = HTTPHandler(client=transport) - if error_type is not None: - with pytest.raises(error_type): + if status == -2: + rejected_params: Final = {**BASE_PARAMS, "api_base": str(body["api_base"])} + if asynchronous: + with pytest.raises(litellm.BadRequestError, match="requires HTTPS"): + await litellm.vector_stores.asearch( + vector_store_id="policy_index", + query="travel policy", + custom_llm_provider="mongodb", + _direct_vector_store_embedding_executor=executor, + **rejected_params, + ) + else: + with pytest.raises(litellm.BadRequestError, match="requires HTTPS"): litellm.vector_stores.search( vector_store_id="policy_index", query="travel policy", custom_llm_provider="mongodb", _direct_vector_store_embedding_executor=executor, - client=client, - timeout=0.75, - **BASE_PARAMS, + **rejected_params, ) - else: - result: Final = litellm.vector_stores.search( - vector_store_id="policy_index", - query="travel policy", - custom_llm_provider="mongodb", - _direct_vector_store_embedding_executor=executor, - client=client, - timeout=0.75, - **BASE_PARAMS, - ) - assert result == RESULT + executor.call.assert_not_called() + return + + def respond(request: httpx.Request) -> httpx.Response: + assert request.url == f"{api_base}/v1/vector_stores/policy_index/search" + assert request.headers["authorization"] == "Bearer test-sidecar-key" + assert request.extensions["timeout"]["read"] == timeout + payload: Final = json.loads(request.content) + assert payload["timeout_ms"] == int(timeout * 1000) + assert payload["query_vector"] == [0.1, 0.2, 0.3] + if status == 0: + raise httpx.ReadTimeout("timed out", request=request) + return httpx.Response(status, json=body) + + async with httpx.AsyncClient(transport=httpx.MockTransport(respond)) as async_transport: + with httpx.Client(transport=httpx.MockTransport(respond)) as transport: + client: Final = AsyncHTTPHandler() if asynchronous else HTTPHandler(client=transport) + if isinstance(client, AsyncHTTPHandler): + await client.client.aclose() + client.client = async_transport + + async def search() -> VectorStoreSearchResponse: + kwargs: Final = { + **BASE_PARAMS, + "api_base": api_base, + "vector_store_id": "policy_index", + "query": "travel policy", + "custom_llm_provider": "mongodb", + "_direct_vector_store_embedding_executor": executor, + "client": client, + "timeout": timeout, + } + if asynchronous: + return await litellm.vector_stores.asearch(**kwargs) + return litellm.vector_stores.search(**kwargs) + + if error_type is not None: + with pytest.raises(error_type): + await search() + else: + assert await search() == RESULT executor.call.assert_called_once_with("embedding-alias", "travel policy", {}) diff --git a/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreForm.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreForm.test.tsx index 8da7098b695..03020bbc4a5 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreForm.test.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreForm.test.tsx @@ -69,7 +69,7 @@ describe("VectorStoreForm", () => { }); }); -const MONGODB_SIDECAR_URL = "http://mongodb-sidecar:8080"; +const MONGODB_SIDECAR_URL = "http://127.0.0.1:8080"; const MONGODB_REQUIRED_FORM_VALUES = { api_base: MONGODB_SIDECAR_URL, diff --git a/ui/litellm-dashboard/src/components/vector_store_providers.tsx b/ui/litellm-dashboard/src/components/vector_store_providers.tsx index 6a8b2f405d2..79e711f6f98 100644 --- a/ui/litellm-dashboard/src/components/vector_store_providers.tsx +++ b/ui/litellm-dashboard/src/components/vector_store_providers.tsx @@ -177,8 +177,8 @@ export const vectorStoreProviderFields: Record { name: "api_base", label: "Sidecar URL", - tooltip: "The URL of your separately deployed MongoDB sidecar. Configure MongoDB credentials in the sidecar", - placeholder: "http://mongodb-sidecar:8080", + tooltip: "Use HTTPS for a remote sidecar, or HTTP with a loopback IP for a sidecar on the same host or Pod", + placeholder: "http://127.0.0.1:8080", required: true, type: "text", },