fix(etl-uvicorn): do not require a body when every input field is optional

CHANGELOG.md

@@ -1,3 +1,19 @@
+## 0.0.45
+
+* **`/invoke` no longer demands a body from a plugin whose parameters are all optional.** A pydantic
+  body parameter with no default is mandatory even when every field inside the model is optional, so
+  such a plugin required a body that no caller has a reason to populate — and before it grew those
+  parameters the same plugin accepted no body at all, which made adding one look
+  backward-compatible while silently flipping the HTTP contract to 422 for every bodyless caller.
+  An absent body now resolves each field to its own default, which is what the signature already
+  promised. Plugins with at least one required field are unchanged: a missing `file_data` still
+  fails validation rather than arriving as `None`.
+* **An optional `file_data` no longer 500s when absent.** The wrapper converted `file_data` from its
+  dict form unconditionally, so a plugin declaring it optional hit
+  `AttributeError: 'NoneType' object has no attribute 'model_dump'` on any body that omitted it —
+  previously reachable via `POST {}`, and via a bodyless request once the change above landed. `None`
+  is now passed through untouched and only a real value is converted.
+
 ## 0.0.44
 
 * **Ignore SIGTERM in plugin uvicorn Servers**: plugin webservers now keep

test/api/test_api.py

@@ -470,3 +470,114 @@ def test_streaming_unstructured_ingest_error_with_none_status_code():
         "Async gen test UnstructuredIngestError with None status_code"
         in invoke_response.status_code_text
     )
+
+
+# --- optional-body contract -------------------------------------------------------------------
+#
+# A pydantic body parameter with no default is mandatory even when every field inside the model is
+# optional. A plugin whose parameters are ALL optional therefore used to demand a body that no
+# caller has a reason to populate: before it grew those parameters the same plugin accepted no body
+# at all, so adding one silently flipped its HTTP contract and every bodyless caller got a 422.
+
+
+class _Echo(BaseModel):
+    settings: Optional[dict] = None
+    context: Optional[dict] = None
+    received: Optional[str] = None
+
+
+def _all_optional(
+    invocation_settings: Optional[dict] = None, invocation_context: Optional[dict] = None
+) -> _Echo:
+    return _Echo(settings=invocation_settings, context=invocation_context)
+
+
+def _no_params() -> _Echo:
+    return _Echo(received="ok")
+
+
+def _has_required(element_dicts: str, invocation_context: Optional[dict] = None) -> _Echo:
+    return _Echo(received=element_dicts)
+
+
+@pytest.mark.parametrize("body", [None, {}])
+def test_all_optional_params_accept_absent_or_empty_body(body):
+    client = TestClient(wrap_in_fastapi(func=_all_optional, plugin_id="mock_plugin"))
+
+    kwargs = {} if body is None else {"json": body}
+    resp = client.post("/invoke", **kwargs)
+
+    assert resp.status_code == 200
+    invoke_response = InvokeResponse.model_validate(resp.json())
+    invoke_response.generic_validation()
+    # Each field resolves to its own default, which is what the signature already promised.
+    assert invoke_response.output == {"settings": None, "context": None, "received": None}
+
+
+def test_all_optional_params_still_receive_a_populated_body():
+    # The tolerance must not swallow a body that IS supplied, or the wire-settings plane silently
+    # stops working while every request keeps returning 200.
+    client = TestClient(wrap_in_fastapi(func=_all_optional, plugin_id="mock_plugin"))
+
+    resp = client.post("/invoke", json={"invocation_settings": {"k": "v"}})
+
+    assert resp.status_code == 200
+    assert InvokeResponse.model_validate(resp.json()).output == {
+        "settings": {"k": "v"},
+        "context": None,
+        "received": None,
+    }
+
+
+def test_required_param_still_rejects_an_absent_body():
+    # The tolerance must not leak into plugins that genuinely need input: a downloader invoked
+    # without `file_data` has to fail loudly rather than receive None.
+    client = TestClient(wrap_in_fastapi(func=_has_required, plugin_id="mock_plugin"))
+
+    assert client.post("/invoke").status_code == 422
+    assert client.post("/invoke", json={}).status_code == 422
+    assert client.post("/invoke", json={"element_dicts": "x"}).status_code == 200
+
+
+class _FileDataEcho(BaseModel):
+    identifier: Optional[str] = None
+
+
+def _optional_file_data(
+    file_data: Optional[FileData] = None, invocation_context: Optional[dict] = None
+) -> _FileDataEcho:
+    return _FileDataEcho(identifier=None if file_data is None else file_data.identifier)
+
+
+@pytest.mark.parametrize("body", [None, {}])
+def test_optional_file_data_is_preserved_as_none(body):
+    # `file_data` is converted from its dict form for the wrapped function, but it can legitimately
+    # be absent. Calling `.model_dump()` on None raised before `wrap_fn` ran, turning the
+    # optional-body contract into a 500 rather than a normal response.
+    client = TestClient(wrap_in_fastapi(func=_optional_file_data, plugin_id="mock_plugin"))
+
+    kwargs = {} if body is None else {"json": body}
+    resp = client.post("/invoke", **kwargs)
+
+    assert resp.status_code == 200
+    invoke_response = InvokeResponse.model_validate(resp.json())
+    invoke_response.generic_validation()
+    assert invoke_response.output == {"identifier": None}
+
+
+def test_optional_file_data_is_still_converted_when_supplied():
+    client = TestClient(wrap_in_fastapi(func=_optional_file_data, plugin_id="mock_plugin"))
+
+    resp = client.post("/invoke", json={"file_data": mock_file_data[0].model_dump()})
+
+    assert resp.status_code == 200
+    assert InvokeResponse.model_validate(resp.json()).output == {"identifier": "mock file data"}
+
+
+def test_no_param_plugin_still_accepts_a_bodyless_post():
+    client = TestClient(wrap_in_fastapi(func=_no_params, plugin_id="mock_plugin"))
+
+    resp = client.post("/invoke")
+
+    assert resp.status_code == 200
+    assert InvokeResponse.model_validate(resp.json()).output["received"] == "ok"

unstructured_platform_plugins/__version__.py

@@ -1 +1 @@
-__version__ = "0.0.44"  # pragma: no cover
+__version__ = "0.0.45"  # pragma: no cover

unstructured_platform_plugins/etl_uvicorn/api_generator.py

@@ -254,22 +254,48 @@ async def _stream_response():
                 file_data=request_dict.get("file_data", None),
             )
 
-    if input_schema_model.model_fields:
+    async def run_job_with_body(request: BaseModel) -> ResponseType:
+        log_func_and_body(func=func, body=request.json())
+        # Create dictionary from pydantic model while preserving underlying types
+        request_dict = {f: getattr(request, f) for f in request.model_fields}
+        # Make sure nested classes get instantiated correctly. `file_data` can legitimately be None
+        # -- a plugin may declare it optional, and then an absent or partial body leaves it unset --
+        # so convert only a real value. Calling `.model_dump()` on None would raise before `wrap_fn`
+        # runs, turning the optional-body contract into a 500.
+        file_data = request_dict.get("file_data")
+        if file_data is not None:
+            request_dict["file_data"] = file_data_from_dict(file_data.model_dump())
+        map_inputs(func=func, raw_inputs=request_dict)
+        if logger.level == LOG_LEVELS.get("trace", logging.NOTSET):
+            logger.log(level=logger.level, msg=f"passing inputs to function: {request_dict}")
+        return await wrap_fn(func=func, kwargs=request_dict)
+
+    # A pydantic body parameter with no default is mandatory even when every field inside the model
+    # is optional. So a plugin whose parameters are ALL optional would demand a body that no caller
+    # has a reason to populate -- and before it grew those parameters that same plugin accepted no
+    # body at all, so adding one flips the HTTP contract while looking backward-compatible. Default
+    # the body in that case: an absent body resolves each field to its own default, which is exactly
+    # what the function signature already promises.
+    #
+    # A plugin with at least one required field keeps a mandatory body, so an indexer that needs
+    # `file_data` still fails validation rather than silently receiving None.
+    body_is_optional = input_schema_model.model_fields and not any(
+        field.is_required() for field in input_schema_model.model_fields.values()
+    )
+
+    if body_is_optional:
+
+        @fastapi_app.post("/invoke", response_model=InvokeResponse)
+        async def run_job(request: Optional[input_schema_model] = None) -> ResponseType:
+            return await run_job_with_body(
+                request if request is not None else input_schema_model()
+            )
+
+    elif input_schema_model.model_fields:
 
         @fastapi_app.post("/invoke", response_model=InvokeResponse)
         async def run_job(request: input_schema_model) -> ResponseType:
-            log_func_and_body(func=func, body=request.json())
-            # Create dictionary from pydantic model while preserving underlying types
-            request_dict = {f: getattr(request, f) for f in request.model_fields}
-            # Make sure nested classes get instantiated correctly
-            if "file_data" in request_dict:
-                request_dict["file_data"] = file_data_from_dict(
-                    request_dict["file_data"].model_dump()
-                )
-            map_inputs(func=func, raw_inputs=request_dict)
-            if logger.level == LOG_LEVELS.get("trace", logging.NOTSET):
-                logger.log(level=logger.level, msg=f"passing inputs to function: {request_dict}")
-            return await wrap_fn(func=func, kwargs=request_dict)
+            return await run_job_with_body(request)
 
     else: