openai/openai-python
Publicmirrored fromhttps://github.com/openai/openai-pythonAvailable
tests/test_client.py
1831lines · modecode
| 1 | # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details. |
| 2 | |
| 3 | from __future__ import annotations |
| 4 | |
| 5 | import gc |
| 6 | import os |
| 7 | import sys |
| 8 | import json |
| 9 | import time |
| 10 | import asyncio |
| 11 | import inspect |
| 12 | import subprocess |
| 13 | import tracemalloc |
| 14 | from typing import Any, Union, cast |
| 15 | from textwrap import dedent |
| 16 | from unittest import mock |
| 17 | from typing_extensions import Literal |
| 18 | |
| 19 | import httpx |
| 20 | import pytest |
| 21 | from respx import MockRouter |
| 22 | from pydantic import ValidationError |
| 23 | |
| 24 | from openai import OpenAI, AsyncOpenAI, APIResponseValidationError |
| 25 | from openai._types import Omit |
| 26 | from openai._utils import maybe_transform |
| 27 | from openai._models import BaseModel, FinalRequestOptions |
| 28 | from openai._constants import RAW_RESPONSE_HEADER |
| 29 | from openai._streaming import Stream, AsyncStream |
| 30 | from openai._exceptions import OpenAIError, APIStatusError, APITimeoutError, APIResponseValidationError |
| 31 | from openai._base_client import DEFAULT_TIMEOUT, HTTPX_DEFAULT_TIMEOUT, BaseClient, make_request_options |
| 32 | from openai.types.chat.completion_create_params import CompletionCreateParamsNonStreaming |
| 33 | |
| 34 | from .utils import update_env |
| 35 | |
| 36 | base_url = os.environ.get("TEST_API_BASE_URL", "http://127.0.0.1:4010") |
| 37 | api_key = "My API Key" |
| 38 | |
| 39 | |
| 40 | def _get_params(client: BaseClient[Any, Any]) -> dict[str, str]: |
| 41 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 42 | url = httpx.URL(request.url) |
| 43 | return dict(url.params) |
| 44 | |
| 45 | |
| 46 | def _low_retry_timeout(*_args: Any, **_kwargs: Any) -> float: |
| 47 | return 0.1 |
| 48 | |
| 49 | |
| 50 | def _get_open_connections(client: OpenAI | AsyncOpenAI) -> int: |
| 51 | transport = client._client._transport |
| 52 | assert isinstance(transport, httpx.HTTPTransport) or isinstance(transport, httpx.AsyncHTTPTransport) |
| 53 | |
| 54 | pool = transport._pool |
| 55 | return len(pool._requests) |
| 56 | |
| 57 | |
| 58 | class TestOpenAI: |
| 59 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 60 | |
| 61 | @pytest.mark.respx(base_url=base_url) |
| 62 | def test_raw_response(self, respx_mock: MockRouter) -> None: |
| 63 | respx_mock.post("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 64 | |
| 65 | response = self.client.post("/foo", cast_to=httpx.Response) |
| 66 | assert response.status_code == 200 |
| 67 | assert isinstance(response, httpx.Response) |
| 68 | assert response.json() == {"foo": "bar"} |
| 69 | |
| 70 | @pytest.mark.respx(base_url=base_url) |
| 71 | def test_raw_response_for_binary(self, respx_mock: MockRouter) -> None: |
| 72 | respx_mock.post("/foo").mock( |
| 73 | return_value=httpx.Response(200, headers={"Content-Type": "application/binary"}, content='{"foo": "bar"}') |
| 74 | ) |
| 75 | |
| 76 | response = self.client.post("/foo", cast_to=httpx.Response) |
| 77 | assert response.status_code == 200 |
| 78 | assert isinstance(response, httpx.Response) |
| 79 | assert response.json() == {"foo": "bar"} |
| 80 | |
| 81 | def test_copy(self) -> None: |
| 82 | copied = self.client.copy() |
| 83 | assert id(copied) != id(self.client) |
| 84 | |
| 85 | copied = self.client.copy(api_key="another My API Key") |
| 86 | assert copied.api_key == "another My API Key" |
| 87 | assert self.client.api_key == "My API Key" |
| 88 | |
| 89 | def test_copy_default_options(self) -> None: |
| 90 | # options that have a default are overridden correctly |
| 91 | copied = self.client.copy(max_retries=7) |
| 92 | assert copied.max_retries == 7 |
| 93 | assert self.client.max_retries == 2 |
| 94 | |
| 95 | copied2 = copied.copy(max_retries=6) |
| 96 | assert copied2.max_retries == 6 |
| 97 | assert copied.max_retries == 7 |
| 98 | |
| 99 | # timeout |
| 100 | assert isinstance(self.client.timeout, httpx.Timeout) |
| 101 | copied = self.client.copy(timeout=None) |
| 102 | assert copied.timeout is None |
| 103 | assert isinstance(self.client.timeout, httpx.Timeout) |
| 104 | |
| 105 | def test_copy_default_headers(self) -> None: |
| 106 | client = OpenAI( |
| 107 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_headers={"X-Foo": "bar"} |
| 108 | ) |
| 109 | assert client.default_headers["X-Foo"] == "bar" |
| 110 | |
| 111 | # does not override the already given value when not specified |
| 112 | copied = client.copy() |
| 113 | assert copied.default_headers["X-Foo"] == "bar" |
| 114 | |
| 115 | # merges already given headers |
| 116 | copied = client.copy(default_headers={"X-Bar": "stainless"}) |
| 117 | assert copied.default_headers["X-Foo"] == "bar" |
| 118 | assert copied.default_headers["X-Bar"] == "stainless" |
| 119 | |
| 120 | # uses new values for any already given headers |
| 121 | copied = client.copy(default_headers={"X-Foo": "stainless"}) |
| 122 | assert copied.default_headers["X-Foo"] == "stainless" |
| 123 | |
| 124 | # set_default_headers |
| 125 | |
| 126 | # completely overrides already set values |
| 127 | copied = client.copy(set_default_headers={}) |
| 128 | assert copied.default_headers.get("X-Foo") is None |
| 129 | |
| 130 | copied = client.copy(set_default_headers={"X-Bar": "Robert"}) |
| 131 | assert copied.default_headers["X-Bar"] == "Robert" |
| 132 | |
| 133 | with pytest.raises( |
| 134 | ValueError, |
| 135 | match="`default_headers` and `set_default_headers` arguments are mutually exclusive", |
| 136 | ): |
| 137 | client.copy(set_default_headers={}, default_headers={"X-Foo": "Bar"}) |
| 138 | |
| 139 | def test_copy_default_query(self) -> None: |
| 140 | client = OpenAI( |
| 141 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_query={"foo": "bar"} |
| 142 | ) |
| 143 | assert _get_params(client)["foo"] == "bar" |
| 144 | |
| 145 | # does not override the already given value when not specified |
| 146 | copied = client.copy() |
| 147 | assert _get_params(copied)["foo"] == "bar" |
| 148 | |
| 149 | # merges already given params |
| 150 | copied = client.copy(default_query={"bar": "stainless"}) |
| 151 | params = _get_params(copied) |
| 152 | assert params["foo"] == "bar" |
| 153 | assert params["bar"] == "stainless" |
| 154 | |
| 155 | # uses new values for any already given headers |
| 156 | copied = client.copy(default_query={"foo": "stainless"}) |
| 157 | assert _get_params(copied)["foo"] == "stainless" |
| 158 | |
| 159 | # set_default_query |
| 160 | |
| 161 | # completely overrides already set values |
| 162 | copied = client.copy(set_default_query={}) |
| 163 | assert _get_params(copied) == {} |
| 164 | |
| 165 | copied = client.copy(set_default_query={"bar": "Robert"}) |
| 166 | assert _get_params(copied)["bar"] == "Robert" |
| 167 | |
| 168 | with pytest.raises( |
| 169 | ValueError, |
| 170 | # TODO: update |
| 171 | match="`default_query` and `set_default_query` arguments are mutually exclusive", |
| 172 | ): |
| 173 | client.copy(set_default_query={}, default_query={"foo": "Bar"}) |
| 174 | |
| 175 | def test_copy_signature(self) -> None: |
| 176 | # ensure the same parameters that can be passed to the client are defined in the `.copy()` method |
| 177 | init_signature = inspect.signature( |
| 178 | # mypy doesn't like that we access the `__init__` property. |
| 179 | self.client.__init__, # type: ignore[misc] |
| 180 | ) |
| 181 | copy_signature = inspect.signature(self.client.copy) |
| 182 | exclude_params = {"transport", "proxies", "_strict_response_validation"} |
| 183 | |
| 184 | for name in init_signature.parameters.keys(): |
| 185 | if name in exclude_params: |
| 186 | continue |
| 187 | |
| 188 | copy_param = copy_signature.parameters.get(name) |
| 189 | assert copy_param is not None, f"copy() signature is missing the {name} param" |
| 190 | |
| 191 | def test_copy_build_request(self) -> None: |
| 192 | options = FinalRequestOptions(method="get", url="/foo") |
| 193 | |
| 194 | def build_request(options: FinalRequestOptions) -> None: |
| 195 | client = self.client.copy() |
| 196 | client._build_request(options) |
| 197 | |
| 198 | # ensure that the machinery is warmed up before tracing starts. |
| 199 | build_request(options) |
| 200 | gc.collect() |
| 201 | |
| 202 | tracemalloc.start(1000) |
| 203 | |
| 204 | snapshot_before = tracemalloc.take_snapshot() |
| 205 | |
| 206 | ITERATIONS = 10 |
| 207 | for _ in range(ITERATIONS): |
| 208 | build_request(options) |
| 209 | |
| 210 | gc.collect() |
| 211 | snapshot_after = tracemalloc.take_snapshot() |
| 212 | |
| 213 | tracemalloc.stop() |
| 214 | |
| 215 | def add_leak(leaks: list[tracemalloc.StatisticDiff], diff: tracemalloc.StatisticDiff) -> None: |
| 216 | if diff.count == 0: |
| 217 | # Avoid false positives by considering only leaks (i.e. allocations that persist). |
| 218 | return |
| 219 | |
| 220 | if diff.count % ITERATIONS != 0: |
| 221 | # Avoid false positives by considering only leaks that appear per iteration. |
| 222 | return |
| 223 | |
| 224 | for frame in diff.traceback: |
| 225 | if any( |
| 226 | frame.filename.endswith(fragment) |
| 227 | for fragment in [ |
| 228 | # to_raw_response_wrapper leaks through the @functools.wraps() decorator. |
| 229 | # |
| 230 | # removing the decorator fixes the leak for reasons we don't understand. |
| 231 | "openai/_legacy_response.py", |
| 232 | "openai/_response.py", |
| 233 | # pydantic.BaseModel.model_dump || pydantic.BaseModel.dict leak memory for some reason. |
| 234 | "openai/_compat.py", |
| 235 | # Standard library leaks we don't care about. |
| 236 | "/logging/__init__.py", |
| 237 | ] |
| 238 | ): |
| 239 | return |
| 240 | |
| 241 | leaks.append(diff) |
| 242 | |
| 243 | leaks: list[tracemalloc.StatisticDiff] = [] |
| 244 | for diff in snapshot_after.compare_to(snapshot_before, "traceback"): |
| 245 | add_leak(leaks, diff) |
| 246 | if leaks: |
| 247 | for leak in leaks: |
| 248 | print("MEMORY LEAK:", leak) |
| 249 | for frame in leak.traceback: |
| 250 | print(frame) |
| 251 | raise AssertionError() |
| 252 | |
| 253 | def test_request_timeout(self) -> None: |
| 254 | request = self.client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 255 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 256 | assert timeout == DEFAULT_TIMEOUT |
| 257 | |
| 258 | request = self.client._build_request( |
| 259 | FinalRequestOptions(method="get", url="/foo", timeout=httpx.Timeout(100.0)) |
| 260 | ) |
| 261 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 262 | assert timeout == httpx.Timeout(100.0) |
| 263 | |
| 264 | def test_client_timeout_option(self) -> None: |
| 265 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True, timeout=httpx.Timeout(0)) |
| 266 | |
| 267 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 268 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 269 | assert timeout == httpx.Timeout(0) |
| 270 | |
| 271 | def test_http_client_timeout_option(self) -> None: |
| 272 | # custom timeout given to the httpx client should be used |
| 273 | with httpx.Client(timeout=None) as http_client: |
| 274 | client = OpenAI( |
| 275 | base_url=base_url, api_key=api_key, _strict_response_validation=True, http_client=http_client |
| 276 | ) |
| 277 | |
| 278 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 279 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 280 | assert timeout == httpx.Timeout(None) |
| 281 | |
| 282 | # no timeout given to the httpx client should not use the httpx default |
| 283 | with httpx.Client() as http_client: |
| 284 | client = OpenAI( |
| 285 | base_url=base_url, api_key=api_key, _strict_response_validation=True, http_client=http_client |
| 286 | ) |
| 287 | |
| 288 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 289 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 290 | assert timeout == DEFAULT_TIMEOUT |
| 291 | |
| 292 | # explicitly passing the default timeout currently results in it being ignored |
| 293 | with httpx.Client(timeout=HTTPX_DEFAULT_TIMEOUT) as http_client: |
| 294 | client = OpenAI( |
| 295 | base_url=base_url, api_key=api_key, _strict_response_validation=True, http_client=http_client |
| 296 | ) |
| 297 | |
| 298 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 299 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 300 | assert timeout == DEFAULT_TIMEOUT # our default |
| 301 | |
| 302 | async def test_invalid_http_client(self) -> None: |
| 303 | with pytest.raises(TypeError, match="Invalid `http_client` arg"): |
| 304 | async with httpx.AsyncClient() as http_client: |
| 305 | OpenAI( |
| 306 | base_url=base_url, |
| 307 | api_key=api_key, |
| 308 | _strict_response_validation=True, |
| 309 | http_client=cast(Any, http_client), |
| 310 | ) |
| 311 | |
| 312 | def test_default_headers_option(self) -> None: |
| 313 | client = OpenAI( |
| 314 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_headers={"X-Foo": "bar"} |
| 315 | ) |
| 316 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 317 | assert request.headers.get("x-foo") == "bar" |
| 318 | assert request.headers.get("x-stainless-lang") == "python" |
| 319 | |
| 320 | client2 = OpenAI( |
| 321 | base_url=base_url, |
| 322 | api_key=api_key, |
| 323 | _strict_response_validation=True, |
| 324 | default_headers={ |
| 325 | "X-Foo": "stainless", |
| 326 | "X-Stainless-Lang": "my-overriding-header", |
| 327 | }, |
| 328 | ) |
| 329 | request = client2._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 330 | assert request.headers.get("x-foo") == "stainless" |
| 331 | assert request.headers.get("x-stainless-lang") == "my-overriding-header" |
| 332 | |
| 333 | def test_validate_headers(self) -> None: |
| 334 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 335 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 336 | assert request.headers.get("Authorization") == f"Bearer {api_key}" |
| 337 | |
| 338 | with pytest.raises(OpenAIError): |
| 339 | with update_env(**{"OPENAI_API_KEY": Omit()}): |
| 340 | client2 = OpenAI(base_url=base_url, api_key=None, _strict_response_validation=True) |
| 341 | _ = client2 |
| 342 | |
| 343 | def test_default_query_option(self) -> None: |
| 344 | client = OpenAI( |
| 345 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_query={"query_param": "bar"} |
| 346 | ) |
| 347 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 348 | url = httpx.URL(request.url) |
| 349 | assert dict(url.params) == {"query_param": "bar"} |
| 350 | |
| 351 | request = client._build_request( |
| 352 | FinalRequestOptions( |
| 353 | method="get", |
| 354 | url="/foo", |
| 355 | params={"foo": "baz", "query_param": "overridden"}, |
| 356 | ) |
| 357 | ) |
| 358 | url = httpx.URL(request.url) |
| 359 | assert dict(url.params) == {"foo": "baz", "query_param": "overridden"} |
| 360 | |
| 361 | def test_request_extra_json(self) -> None: |
| 362 | request = self.client._build_request( |
| 363 | FinalRequestOptions( |
| 364 | method="post", |
| 365 | url="/foo", |
| 366 | json_data={"foo": "bar"}, |
| 367 | extra_json={"baz": False}, |
| 368 | ), |
| 369 | ) |
| 370 | data = json.loads(request.content.decode("utf-8")) |
| 371 | assert data == {"foo": "bar", "baz": False} |
| 372 | |
| 373 | request = self.client._build_request( |
| 374 | FinalRequestOptions( |
| 375 | method="post", |
| 376 | url="/foo", |
| 377 | extra_json={"baz": False}, |
| 378 | ), |
| 379 | ) |
| 380 | data = json.loads(request.content.decode("utf-8")) |
| 381 | assert data == {"baz": False} |
| 382 | |
| 383 | # `extra_json` takes priority over `json_data` when keys clash |
| 384 | request = self.client._build_request( |
| 385 | FinalRequestOptions( |
| 386 | method="post", |
| 387 | url="/foo", |
| 388 | json_data={"foo": "bar", "baz": True}, |
| 389 | extra_json={"baz": None}, |
| 390 | ), |
| 391 | ) |
| 392 | data = json.loads(request.content.decode("utf-8")) |
| 393 | assert data == {"foo": "bar", "baz": None} |
| 394 | |
| 395 | def test_request_extra_headers(self) -> None: |
| 396 | request = self.client._build_request( |
| 397 | FinalRequestOptions( |
| 398 | method="post", |
| 399 | url="/foo", |
| 400 | **make_request_options(extra_headers={"X-Foo": "Foo"}), |
| 401 | ), |
| 402 | ) |
| 403 | assert request.headers.get("X-Foo") == "Foo" |
| 404 | |
| 405 | # `extra_headers` takes priority over `default_headers` when keys clash |
| 406 | request = self.client.with_options(default_headers={"X-Bar": "true"})._build_request( |
| 407 | FinalRequestOptions( |
| 408 | method="post", |
| 409 | url="/foo", |
| 410 | **make_request_options( |
| 411 | extra_headers={"X-Bar": "false"}, |
| 412 | ), |
| 413 | ), |
| 414 | ) |
| 415 | assert request.headers.get("X-Bar") == "false" |
| 416 | |
| 417 | def test_request_extra_query(self) -> None: |
| 418 | request = self.client._build_request( |
| 419 | FinalRequestOptions( |
| 420 | method="post", |
| 421 | url="/foo", |
| 422 | **make_request_options( |
| 423 | extra_query={"my_query_param": "Foo"}, |
| 424 | ), |
| 425 | ), |
| 426 | ) |
| 427 | params = dict(request.url.params) |
| 428 | assert params == {"my_query_param": "Foo"} |
| 429 | |
| 430 | # if both `query` and `extra_query` are given, they are merged |
| 431 | request = self.client._build_request( |
| 432 | FinalRequestOptions( |
| 433 | method="post", |
| 434 | url="/foo", |
| 435 | **make_request_options( |
| 436 | query={"bar": "1"}, |
| 437 | extra_query={"foo": "2"}, |
| 438 | ), |
| 439 | ), |
| 440 | ) |
| 441 | params = dict(request.url.params) |
| 442 | assert params == {"bar": "1", "foo": "2"} |
| 443 | |
| 444 | # `extra_query` takes priority over `query` when keys clash |
| 445 | request = self.client._build_request( |
| 446 | FinalRequestOptions( |
| 447 | method="post", |
| 448 | url="/foo", |
| 449 | **make_request_options( |
| 450 | query={"foo": "1"}, |
| 451 | extra_query={"foo": "2"}, |
| 452 | ), |
| 453 | ), |
| 454 | ) |
| 455 | params = dict(request.url.params) |
| 456 | assert params == {"foo": "2"} |
| 457 | |
| 458 | def test_multipart_repeating_array(self, client: OpenAI) -> None: |
| 459 | request = client._build_request( |
| 460 | FinalRequestOptions.construct( |
| 461 | method="get", |
| 462 | url="/foo", |
| 463 | headers={"Content-Type": "multipart/form-data; boundary=6b7ba517decee4a450543ea6ae821c82"}, |
| 464 | json_data={"array": ["foo", "bar"]}, |
| 465 | files=[("foo.txt", b"hello world")], |
| 466 | ) |
| 467 | ) |
| 468 | |
| 469 | assert request.read().split(b"\r\n") == [ |
| 470 | b"--6b7ba517decee4a450543ea6ae821c82", |
| 471 | b'Content-Disposition: form-data; name="array[]"', |
| 472 | b"", |
| 473 | b"foo", |
| 474 | b"--6b7ba517decee4a450543ea6ae821c82", |
| 475 | b'Content-Disposition: form-data; name="array[]"', |
| 476 | b"", |
| 477 | b"bar", |
| 478 | b"--6b7ba517decee4a450543ea6ae821c82", |
| 479 | b'Content-Disposition: form-data; name="foo.txt"; filename="upload"', |
| 480 | b"Content-Type: application/octet-stream", |
| 481 | b"", |
| 482 | b"hello world", |
| 483 | b"--6b7ba517decee4a450543ea6ae821c82--", |
| 484 | b"", |
| 485 | ] |
| 486 | |
| 487 | @pytest.mark.respx(base_url=base_url) |
| 488 | def test_basic_union_response(self, respx_mock: MockRouter) -> None: |
| 489 | class Model1(BaseModel): |
| 490 | name: str |
| 491 | |
| 492 | class Model2(BaseModel): |
| 493 | foo: str |
| 494 | |
| 495 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 496 | |
| 497 | response = self.client.get("/foo", cast_to=cast(Any, Union[Model1, Model2])) |
| 498 | assert isinstance(response, Model2) |
| 499 | assert response.foo == "bar" |
| 500 | |
| 501 | @pytest.mark.respx(base_url=base_url) |
| 502 | def test_union_response_different_types(self, respx_mock: MockRouter) -> None: |
| 503 | """Union of objects with the same field name using a different type""" |
| 504 | |
| 505 | class Model1(BaseModel): |
| 506 | foo: int |
| 507 | |
| 508 | class Model2(BaseModel): |
| 509 | foo: str |
| 510 | |
| 511 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 512 | |
| 513 | response = self.client.get("/foo", cast_to=cast(Any, Union[Model1, Model2])) |
| 514 | assert isinstance(response, Model2) |
| 515 | assert response.foo == "bar" |
| 516 | |
| 517 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": 1})) |
| 518 | |
| 519 | response = self.client.get("/foo", cast_to=cast(Any, Union[Model1, Model2])) |
| 520 | assert isinstance(response, Model1) |
| 521 | assert response.foo == 1 |
| 522 | |
| 523 | @pytest.mark.respx(base_url=base_url) |
| 524 | def test_non_application_json_content_type_for_json_data(self, respx_mock: MockRouter) -> None: |
| 525 | """ |
| 526 | Response that sets Content-Type to something other than application/json but returns json data |
| 527 | """ |
| 528 | |
| 529 | class Model(BaseModel): |
| 530 | foo: int |
| 531 | |
| 532 | respx_mock.get("/foo").mock( |
| 533 | return_value=httpx.Response( |
| 534 | 200, |
| 535 | content=json.dumps({"foo": 2}), |
| 536 | headers={"Content-Type": "application/text"}, |
| 537 | ) |
| 538 | ) |
| 539 | |
| 540 | response = self.client.get("/foo", cast_to=Model) |
| 541 | assert isinstance(response, Model) |
| 542 | assert response.foo == 2 |
| 543 | |
| 544 | def test_base_url_setter(self) -> None: |
| 545 | client = OpenAI(base_url="https://example.com/from_init", api_key=api_key, _strict_response_validation=True) |
| 546 | assert client.base_url == "https://example.com/from_init/" |
| 547 | |
| 548 | client.base_url = "https://example.com/from_setter" # type: ignore[assignment] |
| 549 | |
| 550 | assert client.base_url == "https://example.com/from_setter/" |
| 551 | |
| 552 | def test_base_url_env(self) -> None: |
| 553 | with update_env(OPENAI_BASE_URL="http://localhost:5000/from/env"): |
| 554 | client = OpenAI(api_key=api_key, _strict_response_validation=True) |
| 555 | assert client.base_url == "http://localhost:5000/from/env/" |
| 556 | |
| 557 | @pytest.mark.parametrize( |
| 558 | "client", |
| 559 | [ |
| 560 | OpenAI(base_url="http://localhost:5000/custom/path/", api_key=api_key, _strict_response_validation=True), |
| 561 | OpenAI( |
| 562 | base_url="http://localhost:5000/custom/path/", |
| 563 | api_key=api_key, |
| 564 | _strict_response_validation=True, |
| 565 | http_client=httpx.Client(), |
| 566 | ), |
| 567 | ], |
| 568 | ids=["standard", "custom http client"], |
| 569 | ) |
| 570 | def test_base_url_trailing_slash(self, client: OpenAI) -> None: |
| 571 | request = client._build_request( |
| 572 | FinalRequestOptions( |
| 573 | method="post", |
| 574 | url="/foo", |
| 575 | json_data={"foo": "bar"}, |
| 576 | ), |
| 577 | ) |
| 578 | assert request.url == "http://localhost:5000/custom/path/foo" |
| 579 | |
| 580 | @pytest.mark.parametrize( |
| 581 | "client", |
| 582 | [ |
| 583 | OpenAI(base_url="http://localhost:5000/custom/path/", api_key=api_key, _strict_response_validation=True), |
| 584 | OpenAI( |
| 585 | base_url="http://localhost:5000/custom/path/", |
| 586 | api_key=api_key, |
| 587 | _strict_response_validation=True, |
| 588 | http_client=httpx.Client(), |
| 589 | ), |
| 590 | ], |
| 591 | ids=["standard", "custom http client"], |
| 592 | ) |
| 593 | def test_base_url_no_trailing_slash(self, client: OpenAI) -> None: |
| 594 | request = client._build_request( |
| 595 | FinalRequestOptions( |
| 596 | method="post", |
| 597 | url="/foo", |
| 598 | json_data={"foo": "bar"}, |
| 599 | ), |
| 600 | ) |
| 601 | assert request.url == "http://localhost:5000/custom/path/foo" |
| 602 | |
| 603 | @pytest.mark.parametrize( |
| 604 | "client", |
| 605 | [ |
| 606 | OpenAI(base_url="http://localhost:5000/custom/path/", api_key=api_key, _strict_response_validation=True), |
| 607 | OpenAI( |
| 608 | base_url="http://localhost:5000/custom/path/", |
| 609 | api_key=api_key, |
| 610 | _strict_response_validation=True, |
| 611 | http_client=httpx.Client(), |
| 612 | ), |
| 613 | ], |
| 614 | ids=["standard", "custom http client"], |
| 615 | ) |
| 616 | def test_absolute_request_url(self, client: OpenAI) -> None: |
| 617 | request = client._build_request( |
| 618 | FinalRequestOptions( |
| 619 | method="post", |
| 620 | url="https://myapi.com/foo", |
| 621 | json_data={"foo": "bar"}, |
| 622 | ), |
| 623 | ) |
| 624 | assert request.url == "https://myapi.com/foo" |
| 625 | |
| 626 | def test_copied_client_does_not_close_http(self) -> None: |
| 627 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 628 | assert not client.is_closed() |
| 629 | |
| 630 | copied = client.copy() |
| 631 | assert copied is not client |
| 632 | |
| 633 | del copied |
| 634 | |
| 635 | assert not client.is_closed() |
| 636 | |
| 637 | def test_client_context_manager(self) -> None: |
| 638 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 639 | with client as c2: |
| 640 | assert c2 is client |
| 641 | assert not c2.is_closed() |
| 642 | assert not client.is_closed() |
| 643 | assert client.is_closed() |
| 644 | |
| 645 | @pytest.mark.respx(base_url=base_url) |
| 646 | def test_client_response_validation_error(self, respx_mock: MockRouter) -> None: |
| 647 | class Model(BaseModel): |
| 648 | foo: str |
| 649 | |
| 650 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": {"invalid": True}})) |
| 651 | |
| 652 | with pytest.raises(APIResponseValidationError) as exc: |
| 653 | self.client.get("/foo", cast_to=Model) |
| 654 | |
| 655 | assert isinstance(exc.value.__cause__, ValidationError) |
| 656 | |
| 657 | def test_client_max_retries_validation(self) -> None: |
| 658 | with pytest.raises(TypeError, match=r"max_retries cannot be None"): |
| 659 | OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True, max_retries=cast(Any, None)) |
| 660 | |
| 661 | @pytest.mark.respx(base_url=base_url) |
| 662 | def test_default_stream_cls(self, respx_mock: MockRouter) -> None: |
| 663 | class Model(BaseModel): |
| 664 | name: str |
| 665 | |
| 666 | respx_mock.post("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 667 | |
| 668 | stream = self.client.post("/foo", cast_to=Model, stream=True, stream_cls=Stream[Model]) |
| 669 | assert isinstance(stream, Stream) |
| 670 | stream.response.close() |
| 671 | |
| 672 | @pytest.mark.respx(base_url=base_url) |
| 673 | def test_received_text_for_expected_json(self, respx_mock: MockRouter) -> None: |
| 674 | class Model(BaseModel): |
| 675 | name: str |
| 676 | |
| 677 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, text="my-custom-format")) |
| 678 | |
| 679 | strict_client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 680 | |
| 681 | with pytest.raises(APIResponseValidationError): |
| 682 | strict_client.get("/foo", cast_to=Model) |
| 683 | |
| 684 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=False) |
| 685 | |
| 686 | response = client.get("/foo", cast_to=Model) |
| 687 | assert isinstance(response, str) # type: ignore[unreachable] |
| 688 | |
| 689 | @pytest.mark.parametrize( |
| 690 | "remaining_retries,retry_after,timeout", |
| 691 | [ |
| 692 | [3, "20", 20], |
| 693 | [3, "0", 0.5], |
| 694 | [3, "-10", 0.5], |
| 695 | [3, "60", 60], |
| 696 | [3, "61", 0.5], |
| 697 | [3, "Fri, 29 Sep 2023 16:26:57 GMT", 20], |
| 698 | [3, "Fri, 29 Sep 2023 16:26:37 GMT", 0.5], |
| 699 | [3, "Fri, 29 Sep 2023 16:26:27 GMT", 0.5], |
| 700 | [3, "Fri, 29 Sep 2023 16:27:37 GMT", 60], |
| 701 | [3, "Fri, 29 Sep 2023 16:27:38 GMT", 0.5], |
| 702 | [3, "99999999999999999999999999999999999", 0.5], |
| 703 | [3, "Zun, 29 Sep 2023 16:26:27 GMT", 0.5], |
| 704 | [3, "", 0.5], |
| 705 | [2, "", 0.5 * 2.0], |
| 706 | [1, "", 0.5 * 4.0], |
| 707 | [-1100, "", 8], # test large number potentially overflowing |
| 708 | ], |
| 709 | ) |
| 710 | @mock.patch("time.time", mock.MagicMock(return_value=1696004797)) |
| 711 | def test_parse_retry_after_header(self, remaining_retries: int, retry_after: str, timeout: float) -> None: |
| 712 | client = OpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 713 | |
| 714 | headers = httpx.Headers({"retry-after": retry_after}) |
| 715 | options = FinalRequestOptions(method="get", url="/foo", max_retries=3) |
| 716 | calculated = client._calculate_retry_timeout(remaining_retries, options, headers) |
| 717 | assert calculated == pytest.approx(timeout, 0.5 * 0.875) # pyright: ignore[reportUnknownMemberType] |
| 718 | |
| 719 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 720 | @pytest.mark.respx(base_url=base_url) |
| 721 | def test_retrying_timeout_errors_doesnt_leak(self, respx_mock: MockRouter) -> None: |
| 722 | respx_mock.post("/chat/completions").mock(side_effect=httpx.TimeoutException("Test timeout error")) |
| 723 | |
| 724 | with pytest.raises(APITimeoutError): |
| 725 | self.client.post( |
| 726 | "/chat/completions", |
| 727 | body=cast( |
| 728 | object, |
| 729 | maybe_transform( |
| 730 | dict( |
| 731 | messages=[ |
| 732 | { |
| 733 | "role": "user", |
| 734 | "content": "Say this is a test", |
| 735 | } |
| 736 | ], |
| 737 | model="gpt-4o", |
| 738 | ), |
| 739 | CompletionCreateParamsNonStreaming, |
| 740 | ), |
| 741 | ), |
| 742 | cast_to=httpx.Response, |
| 743 | options={"headers": {RAW_RESPONSE_HEADER: "stream"}}, |
| 744 | ) |
| 745 | |
| 746 | assert _get_open_connections(self.client) == 0 |
| 747 | |
| 748 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 749 | @pytest.mark.respx(base_url=base_url) |
| 750 | def test_retrying_status_errors_doesnt_leak(self, respx_mock: MockRouter) -> None: |
| 751 | respx_mock.post("/chat/completions").mock(return_value=httpx.Response(500)) |
| 752 | |
| 753 | with pytest.raises(APIStatusError): |
| 754 | self.client.post( |
| 755 | "/chat/completions", |
| 756 | body=cast( |
| 757 | object, |
| 758 | maybe_transform( |
| 759 | dict( |
| 760 | messages=[ |
| 761 | { |
| 762 | "role": "user", |
| 763 | "content": "Say this is a test", |
| 764 | } |
| 765 | ], |
| 766 | model="gpt-4o", |
| 767 | ), |
| 768 | CompletionCreateParamsNonStreaming, |
| 769 | ), |
| 770 | ), |
| 771 | cast_to=httpx.Response, |
| 772 | options={"headers": {RAW_RESPONSE_HEADER: "stream"}}, |
| 773 | ) |
| 774 | |
| 775 | assert _get_open_connections(self.client) == 0 |
| 776 | |
| 777 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 778 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 779 | @pytest.mark.respx(base_url=base_url) |
| 780 | @pytest.mark.parametrize("failure_mode", ["status", "exception"]) |
| 781 | def test_retries_taken( |
| 782 | self, |
| 783 | client: OpenAI, |
| 784 | failures_before_success: int, |
| 785 | failure_mode: Literal["status", "exception"], |
| 786 | respx_mock: MockRouter, |
| 787 | ) -> None: |
| 788 | client = client.with_options(max_retries=4) |
| 789 | |
| 790 | nb_retries = 0 |
| 791 | |
| 792 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 793 | nonlocal nb_retries |
| 794 | if nb_retries < failures_before_success: |
| 795 | nb_retries += 1 |
| 796 | if failure_mode == "exception": |
| 797 | raise RuntimeError("oops") |
| 798 | return httpx.Response(500) |
| 799 | return httpx.Response(200) |
| 800 | |
| 801 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 802 | |
| 803 | response = client.chat.completions.with_raw_response.create( |
| 804 | messages=[ |
| 805 | { |
| 806 | "content": "string", |
| 807 | "role": "developer", |
| 808 | } |
| 809 | ], |
| 810 | model="gpt-4o", |
| 811 | ) |
| 812 | |
| 813 | assert response.retries_taken == failures_before_success |
| 814 | assert int(response.http_request.headers.get("x-stainless-retry-count")) == failures_before_success |
| 815 | |
| 816 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 817 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 818 | @pytest.mark.respx(base_url=base_url) |
| 819 | def test_omit_retry_count_header( |
| 820 | self, client: OpenAI, failures_before_success: int, respx_mock: MockRouter |
| 821 | ) -> None: |
| 822 | client = client.with_options(max_retries=4) |
| 823 | |
| 824 | nb_retries = 0 |
| 825 | |
| 826 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 827 | nonlocal nb_retries |
| 828 | if nb_retries < failures_before_success: |
| 829 | nb_retries += 1 |
| 830 | return httpx.Response(500) |
| 831 | return httpx.Response(200) |
| 832 | |
| 833 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 834 | |
| 835 | response = client.chat.completions.with_raw_response.create( |
| 836 | messages=[ |
| 837 | { |
| 838 | "content": "string", |
| 839 | "role": "developer", |
| 840 | } |
| 841 | ], |
| 842 | model="gpt-4o", |
| 843 | extra_headers={"x-stainless-retry-count": Omit()}, |
| 844 | ) |
| 845 | |
| 846 | assert len(response.http_request.headers.get_list("x-stainless-retry-count")) == 0 |
| 847 | |
| 848 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 849 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 850 | @pytest.mark.respx(base_url=base_url) |
| 851 | def test_overwrite_retry_count_header( |
| 852 | self, client: OpenAI, failures_before_success: int, respx_mock: MockRouter |
| 853 | ) -> None: |
| 854 | client = client.with_options(max_retries=4) |
| 855 | |
| 856 | nb_retries = 0 |
| 857 | |
| 858 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 859 | nonlocal nb_retries |
| 860 | if nb_retries < failures_before_success: |
| 861 | nb_retries += 1 |
| 862 | return httpx.Response(500) |
| 863 | return httpx.Response(200) |
| 864 | |
| 865 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 866 | |
| 867 | response = client.chat.completions.with_raw_response.create( |
| 868 | messages=[ |
| 869 | { |
| 870 | "content": "string", |
| 871 | "role": "developer", |
| 872 | } |
| 873 | ], |
| 874 | model="gpt-4o", |
| 875 | extra_headers={"x-stainless-retry-count": "42"}, |
| 876 | ) |
| 877 | |
| 878 | assert response.http_request.headers.get("x-stainless-retry-count") == "42" |
| 879 | |
| 880 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 881 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 882 | @pytest.mark.respx(base_url=base_url) |
| 883 | def test_retries_taken_new_response_class( |
| 884 | self, client: OpenAI, failures_before_success: int, respx_mock: MockRouter |
| 885 | ) -> None: |
| 886 | client = client.with_options(max_retries=4) |
| 887 | |
| 888 | nb_retries = 0 |
| 889 | |
| 890 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 891 | nonlocal nb_retries |
| 892 | if nb_retries < failures_before_success: |
| 893 | nb_retries += 1 |
| 894 | return httpx.Response(500) |
| 895 | return httpx.Response(200) |
| 896 | |
| 897 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 898 | |
| 899 | with client.chat.completions.with_streaming_response.create( |
| 900 | messages=[ |
| 901 | { |
| 902 | "content": "string", |
| 903 | "role": "developer", |
| 904 | } |
| 905 | ], |
| 906 | model="gpt-4o", |
| 907 | ) as response: |
| 908 | assert response.retries_taken == failures_before_success |
| 909 | assert int(response.http_request.headers.get("x-stainless-retry-count")) == failures_before_success |
| 910 | |
| 911 | |
| 912 | class TestAsyncOpenAI: |
| 913 | client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 914 | |
| 915 | @pytest.mark.respx(base_url=base_url) |
| 916 | @pytest.mark.asyncio |
| 917 | async def test_raw_response(self, respx_mock: MockRouter) -> None: |
| 918 | respx_mock.post("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 919 | |
| 920 | response = await self.client.post("/foo", cast_to=httpx.Response) |
| 921 | assert response.status_code == 200 |
| 922 | assert isinstance(response, httpx.Response) |
| 923 | assert response.json() == {"foo": "bar"} |
| 924 | |
| 925 | @pytest.mark.respx(base_url=base_url) |
| 926 | @pytest.mark.asyncio |
| 927 | async def test_raw_response_for_binary(self, respx_mock: MockRouter) -> None: |
| 928 | respx_mock.post("/foo").mock( |
| 929 | return_value=httpx.Response(200, headers={"Content-Type": "application/binary"}, content='{"foo": "bar"}') |
| 930 | ) |
| 931 | |
| 932 | response = await self.client.post("/foo", cast_to=httpx.Response) |
| 933 | assert response.status_code == 200 |
| 934 | assert isinstance(response, httpx.Response) |
| 935 | assert response.json() == {"foo": "bar"} |
| 936 | |
| 937 | def test_copy(self) -> None: |
| 938 | copied = self.client.copy() |
| 939 | assert id(copied) != id(self.client) |
| 940 | |
| 941 | copied = self.client.copy(api_key="another My API Key") |
| 942 | assert copied.api_key == "another My API Key" |
| 943 | assert self.client.api_key == "My API Key" |
| 944 | |
| 945 | def test_copy_default_options(self) -> None: |
| 946 | # options that have a default are overridden correctly |
| 947 | copied = self.client.copy(max_retries=7) |
| 948 | assert copied.max_retries == 7 |
| 949 | assert self.client.max_retries == 2 |
| 950 | |
| 951 | copied2 = copied.copy(max_retries=6) |
| 952 | assert copied2.max_retries == 6 |
| 953 | assert copied.max_retries == 7 |
| 954 | |
| 955 | # timeout |
| 956 | assert isinstance(self.client.timeout, httpx.Timeout) |
| 957 | copied = self.client.copy(timeout=None) |
| 958 | assert copied.timeout is None |
| 959 | assert isinstance(self.client.timeout, httpx.Timeout) |
| 960 | |
| 961 | def test_copy_default_headers(self) -> None: |
| 962 | client = AsyncOpenAI( |
| 963 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_headers={"X-Foo": "bar"} |
| 964 | ) |
| 965 | assert client.default_headers["X-Foo"] == "bar" |
| 966 | |
| 967 | # does not override the already given value when not specified |
| 968 | copied = client.copy() |
| 969 | assert copied.default_headers["X-Foo"] == "bar" |
| 970 | |
| 971 | # merges already given headers |
| 972 | copied = client.copy(default_headers={"X-Bar": "stainless"}) |
| 973 | assert copied.default_headers["X-Foo"] == "bar" |
| 974 | assert copied.default_headers["X-Bar"] == "stainless" |
| 975 | |
| 976 | # uses new values for any already given headers |
| 977 | copied = client.copy(default_headers={"X-Foo": "stainless"}) |
| 978 | assert copied.default_headers["X-Foo"] == "stainless" |
| 979 | |
| 980 | # set_default_headers |
| 981 | |
| 982 | # completely overrides already set values |
| 983 | copied = client.copy(set_default_headers={}) |
| 984 | assert copied.default_headers.get("X-Foo") is None |
| 985 | |
| 986 | copied = client.copy(set_default_headers={"X-Bar": "Robert"}) |
| 987 | assert copied.default_headers["X-Bar"] == "Robert" |
| 988 | |
| 989 | with pytest.raises( |
| 990 | ValueError, |
| 991 | match="`default_headers` and `set_default_headers` arguments are mutually exclusive", |
| 992 | ): |
| 993 | client.copy(set_default_headers={}, default_headers={"X-Foo": "Bar"}) |
| 994 | |
| 995 | def test_copy_default_query(self) -> None: |
| 996 | client = AsyncOpenAI( |
| 997 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_query={"foo": "bar"} |
| 998 | ) |
| 999 | assert _get_params(client)["foo"] == "bar" |
| 1000 | |
| 1001 | # does not override the already given value when not specified |
| 1002 | copied = client.copy() |
| 1003 | assert _get_params(copied)["foo"] == "bar" |
| 1004 | |
| 1005 | # merges already given params |
| 1006 | copied = client.copy(default_query={"bar": "stainless"}) |
| 1007 | params = _get_params(copied) |
| 1008 | assert params["foo"] == "bar" |
| 1009 | assert params["bar"] == "stainless" |
| 1010 | |
| 1011 | # uses new values for any already given headers |
| 1012 | copied = client.copy(default_query={"foo": "stainless"}) |
| 1013 | assert _get_params(copied)["foo"] == "stainless" |
| 1014 | |
| 1015 | # set_default_query |
| 1016 | |
| 1017 | # completely overrides already set values |
| 1018 | copied = client.copy(set_default_query={}) |
| 1019 | assert _get_params(copied) == {} |
| 1020 | |
| 1021 | copied = client.copy(set_default_query={"bar": "Robert"}) |
| 1022 | assert _get_params(copied)["bar"] == "Robert" |
| 1023 | |
| 1024 | with pytest.raises( |
| 1025 | ValueError, |
| 1026 | # TODO: update |
| 1027 | match="`default_query` and `set_default_query` arguments are mutually exclusive", |
| 1028 | ): |
| 1029 | client.copy(set_default_query={}, default_query={"foo": "Bar"}) |
| 1030 | |
| 1031 | def test_copy_signature(self) -> None: |
| 1032 | # ensure the same parameters that can be passed to the client are defined in the `.copy()` method |
| 1033 | init_signature = inspect.signature( |
| 1034 | # mypy doesn't like that we access the `__init__` property. |
| 1035 | self.client.__init__, # type: ignore[misc] |
| 1036 | ) |
| 1037 | copy_signature = inspect.signature(self.client.copy) |
| 1038 | exclude_params = {"transport", "proxies", "_strict_response_validation"} |
| 1039 | |
| 1040 | for name in init_signature.parameters.keys(): |
| 1041 | if name in exclude_params: |
| 1042 | continue |
| 1043 | |
| 1044 | copy_param = copy_signature.parameters.get(name) |
| 1045 | assert copy_param is not None, f"copy() signature is missing the {name} param" |
| 1046 | |
| 1047 | def test_copy_build_request(self) -> None: |
| 1048 | options = FinalRequestOptions(method="get", url="/foo") |
| 1049 | |
| 1050 | def build_request(options: FinalRequestOptions) -> None: |
| 1051 | client = self.client.copy() |
| 1052 | client._build_request(options) |
| 1053 | |
| 1054 | # ensure that the machinery is warmed up before tracing starts. |
| 1055 | build_request(options) |
| 1056 | gc.collect() |
| 1057 | |
| 1058 | tracemalloc.start(1000) |
| 1059 | |
| 1060 | snapshot_before = tracemalloc.take_snapshot() |
| 1061 | |
| 1062 | ITERATIONS = 10 |
| 1063 | for _ in range(ITERATIONS): |
| 1064 | build_request(options) |
| 1065 | |
| 1066 | gc.collect() |
| 1067 | snapshot_after = tracemalloc.take_snapshot() |
| 1068 | |
| 1069 | tracemalloc.stop() |
| 1070 | |
| 1071 | def add_leak(leaks: list[tracemalloc.StatisticDiff], diff: tracemalloc.StatisticDiff) -> None: |
| 1072 | if diff.count == 0: |
| 1073 | # Avoid false positives by considering only leaks (i.e. allocations that persist). |
| 1074 | return |
| 1075 | |
| 1076 | if diff.count % ITERATIONS != 0: |
| 1077 | # Avoid false positives by considering only leaks that appear per iteration. |
| 1078 | return |
| 1079 | |
| 1080 | for frame in diff.traceback: |
| 1081 | if any( |
| 1082 | frame.filename.endswith(fragment) |
| 1083 | for fragment in [ |
| 1084 | # to_raw_response_wrapper leaks through the @functools.wraps() decorator. |
| 1085 | # |
| 1086 | # removing the decorator fixes the leak for reasons we don't understand. |
| 1087 | "openai/_legacy_response.py", |
| 1088 | "openai/_response.py", |
| 1089 | # pydantic.BaseModel.model_dump || pydantic.BaseModel.dict leak memory for some reason. |
| 1090 | "openai/_compat.py", |
| 1091 | # Standard library leaks we don't care about. |
| 1092 | "/logging/__init__.py", |
| 1093 | ] |
| 1094 | ): |
| 1095 | return |
| 1096 | |
| 1097 | leaks.append(diff) |
| 1098 | |
| 1099 | leaks: list[tracemalloc.StatisticDiff] = [] |
| 1100 | for diff in snapshot_after.compare_to(snapshot_before, "traceback"): |
| 1101 | add_leak(leaks, diff) |
| 1102 | if leaks: |
| 1103 | for leak in leaks: |
| 1104 | print("MEMORY LEAK:", leak) |
| 1105 | for frame in leak.traceback: |
| 1106 | print(frame) |
| 1107 | raise AssertionError() |
| 1108 | |
| 1109 | async def test_request_timeout(self) -> None: |
| 1110 | request = self.client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1111 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 1112 | assert timeout == DEFAULT_TIMEOUT |
| 1113 | |
| 1114 | request = self.client._build_request( |
| 1115 | FinalRequestOptions(method="get", url="/foo", timeout=httpx.Timeout(100.0)) |
| 1116 | ) |
| 1117 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 1118 | assert timeout == httpx.Timeout(100.0) |
| 1119 | |
| 1120 | async def test_client_timeout_option(self) -> None: |
| 1121 | client = AsyncOpenAI( |
| 1122 | base_url=base_url, api_key=api_key, _strict_response_validation=True, timeout=httpx.Timeout(0) |
| 1123 | ) |
| 1124 | |
| 1125 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1126 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 1127 | assert timeout == httpx.Timeout(0) |
| 1128 | |
| 1129 | async def test_http_client_timeout_option(self) -> None: |
| 1130 | # custom timeout given to the httpx client should be used |
| 1131 | async with httpx.AsyncClient(timeout=None) as http_client: |
| 1132 | client = AsyncOpenAI( |
| 1133 | base_url=base_url, api_key=api_key, _strict_response_validation=True, http_client=http_client |
| 1134 | ) |
| 1135 | |
| 1136 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1137 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 1138 | assert timeout == httpx.Timeout(None) |
| 1139 | |
| 1140 | # no timeout given to the httpx client should not use the httpx default |
| 1141 | async with httpx.AsyncClient() as http_client: |
| 1142 | client = AsyncOpenAI( |
| 1143 | base_url=base_url, api_key=api_key, _strict_response_validation=True, http_client=http_client |
| 1144 | ) |
| 1145 | |
| 1146 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1147 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 1148 | assert timeout == DEFAULT_TIMEOUT |
| 1149 | |
| 1150 | # explicitly passing the default timeout currently results in it being ignored |
| 1151 | async with httpx.AsyncClient(timeout=HTTPX_DEFAULT_TIMEOUT) as http_client: |
| 1152 | client = AsyncOpenAI( |
| 1153 | base_url=base_url, api_key=api_key, _strict_response_validation=True, http_client=http_client |
| 1154 | ) |
| 1155 | |
| 1156 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1157 | timeout = httpx.Timeout(**request.extensions["timeout"]) # type: ignore |
| 1158 | assert timeout == DEFAULT_TIMEOUT # our default |
| 1159 | |
| 1160 | def test_invalid_http_client(self) -> None: |
| 1161 | with pytest.raises(TypeError, match="Invalid `http_client` arg"): |
| 1162 | with httpx.Client() as http_client: |
| 1163 | AsyncOpenAI( |
| 1164 | base_url=base_url, |
| 1165 | api_key=api_key, |
| 1166 | _strict_response_validation=True, |
| 1167 | http_client=cast(Any, http_client), |
| 1168 | ) |
| 1169 | |
| 1170 | def test_default_headers_option(self) -> None: |
| 1171 | client = AsyncOpenAI( |
| 1172 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_headers={"X-Foo": "bar"} |
| 1173 | ) |
| 1174 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1175 | assert request.headers.get("x-foo") == "bar" |
| 1176 | assert request.headers.get("x-stainless-lang") == "python" |
| 1177 | |
| 1178 | client2 = AsyncOpenAI( |
| 1179 | base_url=base_url, |
| 1180 | api_key=api_key, |
| 1181 | _strict_response_validation=True, |
| 1182 | default_headers={ |
| 1183 | "X-Foo": "stainless", |
| 1184 | "X-Stainless-Lang": "my-overriding-header", |
| 1185 | }, |
| 1186 | ) |
| 1187 | request = client2._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1188 | assert request.headers.get("x-foo") == "stainless" |
| 1189 | assert request.headers.get("x-stainless-lang") == "my-overriding-header" |
| 1190 | |
| 1191 | def test_validate_headers(self) -> None: |
| 1192 | client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 1193 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1194 | assert request.headers.get("Authorization") == f"Bearer {api_key}" |
| 1195 | |
| 1196 | with pytest.raises(OpenAIError): |
| 1197 | with update_env(**{"OPENAI_API_KEY": Omit()}): |
| 1198 | client2 = AsyncOpenAI(base_url=base_url, api_key=None, _strict_response_validation=True) |
| 1199 | _ = client2 |
| 1200 | |
| 1201 | def test_default_query_option(self) -> None: |
| 1202 | client = AsyncOpenAI( |
| 1203 | base_url=base_url, api_key=api_key, _strict_response_validation=True, default_query={"query_param": "bar"} |
| 1204 | ) |
| 1205 | request = client._build_request(FinalRequestOptions(method="get", url="/foo")) |
| 1206 | url = httpx.URL(request.url) |
| 1207 | assert dict(url.params) == {"query_param": "bar"} |
| 1208 | |
| 1209 | request = client._build_request( |
| 1210 | FinalRequestOptions( |
| 1211 | method="get", |
| 1212 | url="/foo", |
| 1213 | params={"foo": "baz", "query_param": "overridden"}, |
| 1214 | ) |
| 1215 | ) |
| 1216 | url = httpx.URL(request.url) |
| 1217 | assert dict(url.params) == {"foo": "baz", "query_param": "overridden"} |
| 1218 | |
| 1219 | def test_request_extra_json(self) -> None: |
| 1220 | request = self.client._build_request( |
| 1221 | FinalRequestOptions( |
| 1222 | method="post", |
| 1223 | url="/foo", |
| 1224 | json_data={"foo": "bar"}, |
| 1225 | extra_json={"baz": False}, |
| 1226 | ), |
| 1227 | ) |
| 1228 | data = json.loads(request.content.decode("utf-8")) |
| 1229 | assert data == {"foo": "bar", "baz": False} |
| 1230 | |
| 1231 | request = self.client._build_request( |
| 1232 | FinalRequestOptions( |
| 1233 | method="post", |
| 1234 | url="/foo", |
| 1235 | extra_json={"baz": False}, |
| 1236 | ), |
| 1237 | ) |
| 1238 | data = json.loads(request.content.decode("utf-8")) |
| 1239 | assert data == {"baz": False} |
| 1240 | |
| 1241 | # `extra_json` takes priority over `json_data` when keys clash |
| 1242 | request = self.client._build_request( |
| 1243 | FinalRequestOptions( |
| 1244 | method="post", |
| 1245 | url="/foo", |
| 1246 | json_data={"foo": "bar", "baz": True}, |
| 1247 | extra_json={"baz": None}, |
| 1248 | ), |
| 1249 | ) |
| 1250 | data = json.loads(request.content.decode("utf-8")) |
| 1251 | assert data == {"foo": "bar", "baz": None} |
| 1252 | |
| 1253 | def test_request_extra_headers(self) -> None: |
| 1254 | request = self.client._build_request( |
| 1255 | FinalRequestOptions( |
| 1256 | method="post", |
| 1257 | url="/foo", |
| 1258 | **make_request_options(extra_headers={"X-Foo": "Foo"}), |
| 1259 | ), |
| 1260 | ) |
| 1261 | assert request.headers.get("X-Foo") == "Foo" |
| 1262 | |
| 1263 | # `extra_headers` takes priority over `default_headers` when keys clash |
| 1264 | request = self.client.with_options(default_headers={"X-Bar": "true"})._build_request( |
| 1265 | FinalRequestOptions( |
| 1266 | method="post", |
| 1267 | url="/foo", |
| 1268 | **make_request_options( |
| 1269 | extra_headers={"X-Bar": "false"}, |
| 1270 | ), |
| 1271 | ), |
| 1272 | ) |
| 1273 | assert request.headers.get("X-Bar") == "false" |
| 1274 | |
| 1275 | def test_request_extra_query(self) -> None: |
| 1276 | request = self.client._build_request( |
| 1277 | FinalRequestOptions( |
| 1278 | method="post", |
| 1279 | url="/foo", |
| 1280 | **make_request_options( |
| 1281 | extra_query={"my_query_param": "Foo"}, |
| 1282 | ), |
| 1283 | ), |
| 1284 | ) |
| 1285 | params = dict(request.url.params) |
| 1286 | assert params == {"my_query_param": "Foo"} |
| 1287 | |
| 1288 | # if both `query` and `extra_query` are given, they are merged |
| 1289 | request = self.client._build_request( |
| 1290 | FinalRequestOptions( |
| 1291 | method="post", |
| 1292 | url="/foo", |
| 1293 | **make_request_options( |
| 1294 | query={"bar": "1"}, |
| 1295 | extra_query={"foo": "2"}, |
| 1296 | ), |
| 1297 | ), |
| 1298 | ) |
| 1299 | params = dict(request.url.params) |
| 1300 | assert params == {"bar": "1", "foo": "2"} |
| 1301 | |
| 1302 | # `extra_query` takes priority over `query` when keys clash |
| 1303 | request = self.client._build_request( |
| 1304 | FinalRequestOptions( |
| 1305 | method="post", |
| 1306 | url="/foo", |
| 1307 | **make_request_options( |
| 1308 | query={"foo": "1"}, |
| 1309 | extra_query={"foo": "2"}, |
| 1310 | ), |
| 1311 | ), |
| 1312 | ) |
| 1313 | params = dict(request.url.params) |
| 1314 | assert params == {"foo": "2"} |
| 1315 | |
| 1316 | def test_multipart_repeating_array(self, async_client: AsyncOpenAI) -> None: |
| 1317 | request = async_client._build_request( |
| 1318 | FinalRequestOptions.construct( |
| 1319 | method="get", |
| 1320 | url="/foo", |
| 1321 | headers={"Content-Type": "multipart/form-data; boundary=6b7ba517decee4a450543ea6ae821c82"}, |
| 1322 | json_data={"array": ["foo", "bar"]}, |
| 1323 | files=[("foo.txt", b"hello world")], |
| 1324 | ) |
| 1325 | ) |
| 1326 | |
| 1327 | assert request.read().split(b"\r\n") == [ |
| 1328 | b"--6b7ba517decee4a450543ea6ae821c82", |
| 1329 | b'Content-Disposition: form-data; name="array[]"', |
| 1330 | b"", |
| 1331 | b"foo", |
| 1332 | b"--6b7ba517decee4a450543ea6ae821c82", |
| 1333 | b'Content-Disposition: form-data; name="array[]"', |
| 1334 | b"", |
| 1335 | b"bar", |
| 1336 | b"--6b7ba517decee4a450543ea6ae821c82", |
| 1337 | b'Content-Disposition: form-data; name="foo.txt"; filename="upload"', |
| 1338 | b"Content-Type: application/octet-stream", |
| 1339 | b"", |
| 1340 | b"hello world", |
| 1341 | b"--6b7ba517decee4a450543ea6ae821c82--", |
| 1342 | b"", |
| 1343 | ] |
| 1344 | |
| 1345 | @pytest.mark.respx(base_url=base_url) |
| 1346 | async def test_basic_union_response(self, respx_mock: MockRouter) -> None: |
| 1347 | class Model1(BaseModel): |
| 1348 | name: str |
| 1349 | |
| 1350 | class Model2(BaseModel): |
| 1351 | foo: str |
| 1352 | |
| 1353 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 1354 | |
| 1355 | response = await self.client.get("/foo", cast_to=cast(Any, Union[Model1, Model2])) |
| 1356 | assert isinstance(response, Model2) |
| 1357 | assert response.foo == "bar" |
| 1358 | |
| 1359 | @pytest.mark.respx(base_url=base_url) |
| 1360 | async def test_union_response_different_types(self, respx_mock: MockRouter) -> None: |
| 1361 | """Union of objects with the same field name using a different type""" |
| 1362 | |
| 1363 | class Model1(BaseModel): |
| 1364 | foo: int |
| 1365 | |
| 1366 | class Model2(BaseModel): |
| 1367 | foo: str |
| 1368 | |
| 1369 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 1370 | |
| 1371 | response = await self.client.get("/foo", cast_to=cast(Any, Union[Model1, Model2])) |
| 1372 | assert isinstance(response, Model2) |
| 1373 | assert response.foo == "bar" |
| 1374 | |
| 1375 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": 1})) |
| 1376 | |
| 1377 | response = await self.client.get("/foo", cast_to=cast(Any, Union[Model1, Model2])) |
| 1378 | assert isinstance(response, Model1) |
| 1379 | assert response.foo == 1 |
| 1380 | |
| 1381 | @pytest.mark.respx(base_url=base_url) |
| 1382 | async def test_non_application_json_content_type_for_json_data(self, respx_mock: MockRouter) -> None: |
| 1383 | """ |
| 1384 | Response that sets Content-Type to something other than application/json but returns json data |
| 1385 | """ |
| 1386 | |
| 1387 | class Model(BaseModel): |
| 1388 | foo: int |
| 1389 | |
| 1390 | respx_mock.get("/foo").mock( |
| 1391 | return_value=httpx.Response( |
| 1392 | 200, |
| 1393 | content=json.dumps({"foo": 2}), |
| 1394 | headers={"Content-Type": "application/text"}, |
| 1395 | ) |
| 1396 | ) |
| 1397 | |
| 1398 | response = await self.client.get("/foo", cast_to=Model) |
| 1399 | assert isinstance(response, Model) |
| 1400 | assert response.foo == 2 |
| 1401 | |
| 1402 | def test_base_url_setter(self) -> None: |
| 1403 | client = AsyncOpenAI( |
| 1404 | base_url="https://example.com/from_init", api_key=api_key, _strict_response_validation=True |
| 1405 | ) |
| 1406 | assert client.base_url == "https://example.com/from_init/" |
| 1407 | |
| 1408 | client.base_url = "https://example.com/from_setter" # type: ignore[assignment] |
| 1409 | |
| 1410 | assert client.base_url == "https://example.com/from_setter/" |
| 1411 | |
| 1412 | def test_base_url_env(self) -> None: |
| 1413 | with update_env(OPENAI_BASE_URL="http://localhost:5000/from/env"): |
| 1414 | client = AsyncOpenAI(api_key=api_key, _strict_response_validation=True) |
| 1415 | assert client.base_url == "http://localhost:5000/from/env/" |
| 1416 | |
| 1417 | @pytest.mark.parametrize( |
| 1418 | "client", |
| 1419 | [ |
| 1420 | AsyncOpenAI( |
| 1421 | base_url="http://localhost:5000/custom/path/", api_key=api_key, _strict_response_validation=True |
| 1422 | ), |
| 1423 | AsyncOpenAI( |
| 1424 | base_url="http://localhost:5000/custom/path/", |
| 1425 | api_key=api_key, |
| 1426 | _strict_response_validation=True, |
| 1427 | http_client=httpx.AsyncClient(), |
| 1428 | ), |
| 1429 | ], |
| 1430 | ids=["standard", "custom http client"], |
| 1431 | ) |
| 1432 | def test_base_url_trailing_slash(self, client: AsyncOpenAI) -> None: |
| 1433 | request = client._build_request( |
| 1434 | FinalRequestOptions( |
| 1435 | method="post", |
| 1436 | url="/foo", |
| 1437 | json_data={"foo": "bar"}, |
| 1438 | ), |
| 1439 | ) |
| 1440 | assert request.url == "http://localhost:5000/custom/path/foo" |
| 1441 | |
| 1442 | @pytest.mark.parametrize( |
| 1443 | "client", |
| 1444 | [ |
| 1445 | AsyncOpenAI( |
| 1446 | base_url="http://localhost:5000/custom/path/", api_key=api_key, _strict_response_validation=True |
| 1447 | ), |
| 1448 | AsyncOpenAI( |
| 1449 | base_url="http://localhost:5000/custom/path/", |
| 1450 | api_key=api_key, |
| 1451 | _strict_response_validation=True, |
| 1452 | http_client=httpx.AsyncClient(), |
| 1453 | ), |
| 1454 | ], |
| 1455 | ids=["standard", "custom http client"], |
| 1456 | ) |
| 1457 | def test_base_url_no_trailing_slash(self, client: AsyncOpenAI) -> None: |
| 1458 | request = client._build_request( |
| 1459 | FinalRequestOptions( |
| 1460 | method="post", |
| 1461 | url="/foo", |
| 1462 | json_data={"foo": "bar"}, |
| 1463 | ), |
| 1464 | ) |
| 1465 | assert request.url == "http://localhost:5000/custom/path/foo" |
| 1466 | |
| 1467 | @pytest.mark.parametrize( |
| 1468 | "client", |
| 1469 | [ |
| 1470 | AsyncOpenAI( |
| 1471 | base_url="http://localhost:5000/custom/path/", api_key=api_key, _strict_response_validation=True |
| 1472 | ), |
| 1473 | AsyncOpenAI( |
| 1474 | base_url="http://localhost:5000/custom/path/", |
| 1475 | api_key=api_key, |
| 1476 | _strict_response_validation=True, |
| 1477 | http_client=httpx.AsyncClient(), |
| 1478 | ), |
| 1479 | ], |
| 1480 | ids=["standard", "custom http client"], |
| 1481 | ) |
| 1482 | def test_absolute_request_url(self, client: AsyncOpenAI) -> None: |
| 1483 | request = client._build_request( |
| 1484 | FinalRequestOptions( |
| 1485 | method="post", |
| 1486 | url="https://myapi.com/foo", |
| 1487 | json_data={"foo": "bar"}, |
| 1488 | ), |
| 1489 | ) |
| 1490 | assert request.url == "https://myapi.com/foo" |
| 1491 | |
| 1492 | async def test_copied_client_does_not_close_http(self) -> None: |
| 1493 | client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 1494 | assert not client.is_closed() |
| 1495 | |
| 1496 | copied = client.copy() |
| 1497 | assert copied is not client |
| 1498 | |
| 1499 | del copied |
| 1500 | |
| 1501 | await asyncio.sleep(0.2) |
| 1502 | assert not client.is_closed() |
| 1503 | |
| 1504 | async def test_client_context_manager(self) -> None: |
| 1505 | client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 1506 | async with client as c2: |
| 1507 | assert c2 is client |
| 1508 | assert not c2.is_closed() |
| 1509 | assert not client.is_closed() |
| 1510 | assert client.is_closed() |
| 1511 | |
| 1512 | @pytest.mark.respx(base_url=base_url) |
| 1513 | @pytest.mark.asyncio |
| 1514 | async def test_client_response_validation_error(self, respx_mock: MockRouter) -> None: |
| 1515 | class Model(BaseModel): |
| 1516 | foo: str |
| 1517 | |
| 1518 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, json={"foo": {"invalid": True}})) |
| 1519 | |
| 1520 | with pytest.raises(APIResponseValidationError) as exc: |
| 1521 | await self.client.get("/foo", cast_to=Model) |
| 1522 | |
| 1523 | assert isinstance(exc.value.__cause__, ValidationError) |
| 1524 | |
| 1525 | async def test_client_max_retries_validation(self) -> None: |
| 1526 | with pytest.raises(TypeError, match=r"max_retries cannot be None"): |
| 1527 | AsyncOpenAI( |
| 1528 | base_url=base_url, api_key=api_key, _strict_response_validation=True, max_retries=cast(Any, None) |
| 1529 | ) |
| 1530 | |
| 1531 | @pytest.mark.respx(base_url=base_url) |
| 1532 | @pytest.mark.asyncio |
| 1533 | async def test_default_stream_cls(self, respx_mock: MockRouter) -> None: |
| 1534 | class Model(BaseModel): |
| 1535 | name: str |
| 1536 | |
| 1537 | respx_mock.post("/foo").mock(return_value=httpx.Response(200, json={"foo": "bar"})) |
| 1538 | |
| 1539 | stream = await self.client.post("/foo", cast_to=Model, stream=True, stream_cls=AsyncStream[Model]) |
| 1540 | assert isinstance(stream, AsyncStream) |
| 1541 | await stream.response.aclose() |
| 1542 | |
| 1543 | @pytest.mark.respx(base_url=base_url) |
| 1544 | @pytest.mark.asyncio |
| 1545 | async def test_received_text_for_expected_json(self, respx_mock: MockRouter) -> None: |
| 1546 | class Model(BaseModel): |
| 1547 | name: str |
| 1548 | |
| 1549 | respx_mock.get("/foo").mock(return_value=httpx.Response(200, text="my-custom-format")) |
| 1550 | |
| 1551 | strict_client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 1552 | |
| 1553 | with pytest.raises(APIResponseValidationError): |
| 1554 | await strict_client.get("/foo", cast_to=Model) |
| 1555 | |
| 1556 | client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=False) |
| 1557 | |
| 1558 | response = await client.get("/foo", cast_to=Model) |
| 1559 | assert isinstance(response, str) # type: ignore[unreachable] |
| 1560 | |
| 1561 | @pytest.mark.parametrize( |
| 1562 | "remaining_retries,retry_after,timeout", |
| 1563 | [ |
| 1564 | [3, "20", 20], |
| 1565 | [3, "0", 0.5], |
| 1566 | [3, "-10", 0.5], |
| 1567 | [3, "60", 60], |
| 1568 | [3, "61", 0.5], |
| 1569 | [3, "Fri, 29 Sep 2023 16:26:57 GMT", 20], |
| 1570 | [3, "Fri, 29 Sep 2023 16:26:37 GMT", 0.5], |
| 1571 | [3, "Fri, 29 Sep 2023 16:26:27 GMT", 0.5], |
| 1572 | [3, "Fri, 29 Sep 2023 16:27:37 GMT", 60], |
| 1573 | [3, "Fri, 29 Sep 2023 16:27:38 GMT", 0.5], |
| 1574 | [3, "99999999999999999999999999999999999", 0.5], |
| 1575 | [3, "Zun, 29 Sep 2023 16:26:27 GMT", 0.5], |
| 1576 | [3, "", 0.5], |
| 1577 | [2, "", 0.5 * 2.0], |
| 1578 | [1, "", 0.5 * 4.0], |
| 1579 | [-1100, "", 8], # test large number potentially overflowing |
| 1580 | ], |
| 1581 | ) |
| 1582 | @mock.patch("time.time", mock.MagicMock(return_value=1696004797)) |
| 1583 | @pytest.mark.asyncio |
| 1584 | async def test_parse_retry_after_header(self, remaining_retries: int, retry_after: str, timeout: float) -> None: |
| 1585 | client = AsyncOpenAI(base_url=base_url, api_key=api_key, _strict_response_validation=True) |
| 1586 | |
| 1587 | headers = httpx.Headers({"retry-after": retry_after}) |
| 1588 | options = FinalRequestOptions(method="get", url="/foo", max_retries=3) |
| 1589 | calculated = client._calculate_retry_timeout(remaining_retries, options, headers) |
| 1590 | assert calculated == pytest.approx(timeout, 0.5 * 0.875) # pyright: ignore[reportUnknownMemberType] |
| 1591 | |
| 1592 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 1593 | @pytest.mark.respx(base_url=base_url) |
| 1594 | async def test_retrying_timeout_errors_doesnt_leak(self, respx_mock: MockRouter) -> None: |
| 1595 | respx_mock.post("/chat/completions").mock(side_effect=httpx.TimeoutException("Test timeout error")) |
| 1596 | |
| 1597 | with pytest.raises(APITimeoutError): |
| 1598 | await self.client.post( |
| 1599 | "/chat/completions", |
| 1600 | body=cast( |
| 1601 | object, |
| 1602 | maybe_transform( |
| 1603 | dict( |
| 1604 | messages=[ |
| 1605 | { |
| 1606 | "role": "user", |
| 1607 | "content": "Say this is a test", |
| 1608 | } |
| 1609 | ], |
| 1610 | model="gpt-4o", |
| 1611 | ), |
| 1612 | CompletionCreateParamsNonStreaming, |
| 1613 | ), |
| 1614 | ), |
| 1615 | cast_to=httpx.Response, |
| 1616 | options={"headers": {RAW_RESPONSE_HEADER: "stream"}}, |
| 1617 | ) |
| 1618 | |
| 1619 | assert _get_open_connections(self.client) == 0 |
| 1620 | |
| 1621 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 1622 | @pytest.mark.respx(base_url=base_url) |
| 1623 | async def test_retrying_status_errors_doesnt_leak(self, respx_mock: MockRouter) -> None: |
| 1624 | respx_mock.post("/chat/completions").mock(return_value=httpx.Response(500)) |
| 1625 | |
| 1626 | with pytest.raises(APIStatusError): |
| 1627 | await self.client.post( |
| 1628 | "/chat/completions", |
| 1629 | body=cast( |
| 1630 | object, |
| 1631 | maybe_transform( |
| 1632 | dict( |
| 1633 | messages=[ |
| 1634 | { |
| 1635 | "role": "user", |
| 1636 | "content": "Say this is a test", |
| 1637 | } |
| 1638 | ], |
| 1639 | model="gpt-4o", |
| 1640 | ), |
| 1641 | CompletionCreateParamsNonStreaming, |
| 1642 | ), |
| 1643 | ), |
| 1644 | cast_to=httpx.Response, |
| 1645 | options={"headers": {RAW_RESPONSE_HEADER: "stream"}}, |
| 1646 | ) |
| 1647 | |
| 1648 | assert _get_open_connections(self.client) == 0 |
| 1649 | |
| 1650 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 1651 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 1652 | @pytest.mark.respx(base_url=base_url) |
| 1653 | @pytest.mark.asyncio |
| 1654 | @pytest.mark.parametrize("failure_mode", ["status", "exception"]) |
| 1655 | async def test_retries_taken( |
| 1656 | self, |
| 1657 | async_client: AsyncOpenAI, |
| 1658 | failures_before_success: int, |
| 1659 | failure_mode: Literal["status", "exception"], |
| 1660 | respx_mock: MockRouter, |
| 1661 | ) -> None: |
| 1662 | client = async_client.with_options(max_retries=4) |
| 1663 | |
| 1664 | nb_retries = 0 |
| 1665 | |
| 1666 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 1667 | nonlocal nb_retries |
| 1668 | if nb_retries < failures_before_success: |
| 1669 | nb_retries += 1 |
| 1670 | if failure_mode == "exception": |
| 1671 | raise RuntimeError("oops") |
| 1672 | return httpx.Response(500) |
| 1673 | return httpx.Response(200) |
| 1674 | |
| 1675 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 1676 | |
| 1677 | response = await client.chat.completions.with_raw_response.create( |
| 1678 | messages=[ |
| 1679 | { |
| 1680 | "content": "string", |
| 1681 | "role": "developer", |
| 1682 | } |
| 1683 | ], |
| 1684 | model="gpt-4o", |
| 1685 | ) |
| 1686 | |
| 1687 | assert response.retries_taken == failures_before_success |
| 1688 | assert int(response.http_request.headers.get("x-stainless-retry-count")) == failures_before_success |
| 1689 | |
| 1690 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 1691 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 1692 | @pytest.mark.respx(base_url=base_url) |
| 1693 | @pytest.mark.asyncio |
| 1694 | async def test_omit_retry_count_header( |
| 1695 | self, async_client: AsyncOpenAI, failures_before_success: int, respx_mock: MockRouter |
| 1696 | ) -> None: |
| 1697 | client = async_client.with_options(max_retries=4) |
| 1698 | |
| 1699 | nb_retries = 0 |
| 1700 | |
| 1701 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 1702 | nonlocal nb_retries |
| 1703 | if nb_retries < failures_before_success: |
| 1704 | nb_retries += 1 |
| 1705 | return httpx.Response(500) |
| 1706 | return httpx.Response(200) |
| 1707 | |
| 1708 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 1709 | |
| 1710 | response = await client.chat.completions.with_raw_response.create( |
| 1711 | messages=[ |
| 1712 | { |
| 1713 | "content": "string", |
| 1714 | "role": "developer", |
| 1715 | } |
| 1716 | ], |
| 1717 | model="gpt-4o", |
| 1718 | extra_headers={"x-stainless-retry-count": Omit()}, |
| 1719 | ) |
| 1720 | |
| 1721 | assert len(response.http_request.headers.get_list("x-stainless-retry-count")) == 0 |
| 1722 | |
| 1723 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 1724 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 1725 | @pytest.mark.respx(base_url=base_url) |
| 1726 | @pytest.mark.asyncio |
| 1727 | async def test_overwrite_retry_count_header( |
| 1728 | self, async_client: AsyncOpenAI, failures_before_success: int, respx_mock: MockRouter |
| 1729 | ) -> None: |
| 1730 | client = async_client.with_options(max_retries=4) |
| 1731 | |
| 1732 | nb_retries = 0 |
| 1733 | |
| 1734 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 1735 | nonlocal nb_retries |
| 1736 | if nb_retries < failures_before_success: |
| 1737 | nb_retries += 1 |
| 1738 | return httpx.Response(500) |
| 1739 | return httpx.Response(200) |
| 1740 | |
| 1741 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 1742 | |
| 1743 | response = await client.chat.completions.with_raw_response.create( |
| 1744 | messages=[ |
| 1745 | { |
| 1746 | "content": "string", |
| 1747 | "role": "developer", |
| 1748 | } |
| 1749 | ], |
| 1750 | model="gpt-4o", |
| 1751 | extra_headers={"x-stainless-retry-count": "42"}, |
| 1752 | ) |
| 1753 | |
| 1754 | assert response.http_request.headers.get("x-stainless-retry-count") == "42" |
| 1755 | |
| 1756 | @pytest.mark.parametrize("failures_before_success", [0, 2, 4]) |
| 1757 | @mock.patch("openai._base_client.BaseClient._calculate_retry_timeout", _low_retry_timeout) |
| 1758 | @pytest.mark.respx(base_url=base_url) |
| 1759 | @pytest.mark.asyncio |
| 1760 | async def test_retries_taken_new_response_class( |
| 1761 | self, async_client: AsyncOpenAI, failures_before_success: int, respx_mock: MockRouter |
| 1762 | ) -> None: |
| 1763 | client = async_client.with_options(max_retries=4) |
| 1764 | |
| 1765 | nb_retries = 0 |
| 1766 | |
| 1767 | def retry_handler(_request: httpx.Request) -> httpx.Response: |
| 1768 | nonlocal nb_retries |
| 1769 | if nb_retries < failures_before_success: |
| 1770 | nb_retries += 1 |
| 1771 | return httpx.Response(500) |
| 1772 | return httpx.Response(200) |
| 1773 | |
| 1774 | respx_mock.post("/chat/completions").mock(side_effect=retry_handler) |
| 1775 | |
| 1776 | async with client.chat.completions.with_streaming_response.create( |
| 1777 | messages=[ |
| 1778 | { |
| 1779 | "content": "string", |
| 1780 | "role": "developer", |
| 1781 | } |
| 1782 | ], |
| 1783 | model="gpt-4o", |
| 1784 | ) as response: |
| 1785 | assert response.retries_taken == failures_before_success |
| 1786 | assert int(response.http_request.headers.get("x-stainless-retry-count")) == failures_before_success |
| 1787 | |
| 1788 | def test_get_platform(self) -> None: |
| 1789 | # A previous implementation of asyncify could leave threads unterminated when |
| 1790 | # used with nest_asyncio. |
| 1791 | # |
| 1792 | # Since nest_asyncio.apply() is global and cannot be un-applied, this |
| 1793 | # test is run in a separate process to avoid affecting other tests. |
| 1794 | test_code = dedent(""" |
| 1795 | import asyncio |
| 1796 | import nest_asyncio |
| 1797 | import threading |
| 1798 | |
| 1799 | from openai._utils import asyncify |
| 1800 | from openai._base_client import get_platform |
| 1801 | |
| 1802 | async def test_main() -> None: |
| 1803 | result = await asyncify(get_platform)() |
| 1804 | print(result) |
| 1805 | for thread in threading.enumerate(): |
| 1806 | print(thread.name) |
| 1807 | |
| 1808 | nest_asyncio.apply() |
| 1809 | asyncio.run(test_main()) |
| 1810 | """) |
| 1811 | with subprocess.Popen( |
| 1812 | [sys.executable, "-c", test_code], |
| 1813 | text=True, |
| 1814 | ) as process: |
| 1815 | timeout = 10 # seconds |
| 1816 | |
| 1817 | start_time = time.monotonic() |
| 1818 | while True: |
| 1819 | return_code = process.poll() |
| 1820 | if return_code is not None: |
| 1821 | if return_code != 0: |
| 1822 | raise AssertionError("calling get_platform using asyncify resulted in a non-zero exit code") |
| 1823 | |
| 1824 | # success |
| 1825 | break |
| 1826 | |
| 1827 | if time.monotonic() - start_time > timeout: |
| 1828 | process.kill() |
| 1829 | raise AssertionError("calling get_platform using asyncify resulted in a hung process") |
| 1830 | |
| 1831 | time.sleep(0.1) |
| 1832 | |