| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | import asyncio |
| 4 | |
| 5 | import openai |
| 6 | import pytest |
| 7 | |
| 8 | |
| 9 | @pytest.mark.asyncio |
| 10 | async def test_store(client: openai.AsyncOpenAI): |
| 11 | # By default, store is True. |
| 12 | response = await client.responses.create(input="Hello!") |
| 13 | assert response.status == "completed" |
| 14 | |
| 15 | # Retrieve the response. |
| 16 | response = await client.responses.retrieve(response.id) |
| 17 | assert response.status == "completed" |
| 18 | |
| 19 | # Test store=False. |
| 20 | response = await client.responses.create( |
| 21 | input="Hello!", |
| 22 | store=False, |
| 23 | ) |
| 24 | assert response.status == "completed" |
| 25 | |
| 26 | # The response should not be found. |
| 27 | with pytest.raises(openai.NotFoundError, match="Response with id .* not found."): |
| 28 | await client.responses.retrieve(response.id) |
| 29 | |
| 30 | |
| 31 | @pytest.mark.asyncio |
| 32 | async def test_background(client: openai.AsyncOpenAI): |
| 33 | # NOTE: This query should be easy enough for the model to answer |
| 34 | # within the 10 seconds. |
| 35 | response = await client.responses.create( |
| 36 | input="Hello!", |
| 37 | background=True, |
| 38 | ) |
| 39 | assert response.status == "queued" |
| 40 | |
| 41 | max_retries = 10 |
| 42 | for _ in range(max_retries): |
| 43 | await asyncio.sleep(1) |
| 44 | response = await client.responses.retrieve(response.id) |
| 45 | if response.status != "queued": |
| 46 | break |
| 47 | print(response) |
| 48 | |
| 49 | assert response.status == "completed" |
| 50 | |
| 51 | |
| 52 | @pytest.mark.asyncio |
| 53 | async def test_background_error(client: openai.AsyncOpenAI): |
| 54 | with pytest.raises( |
| 55 | openai.BadRequestError, match="background can only be used when `store` is true" |
| 56 | ): |
| 57 | _ = await client.responses.create( |
| 58 | input="What is 13 * 24?", |
| 59 | background=True, |
| 60 | store=False, |
| 61 | ) |
| 62 | |
| 63 | |
| 64 | @pytest.mark.asyncio |
| 65 | async def test_background_cancel(client: openai.AsyncOpenAI): |
| 66 | response = await client.responses.create( |
| 67 | input="Write a long story about a cat.", |
| 68 | background=True, |
| 69 | ) |
| 70 | assert response.status == "queued" |
| 71 | |
| 72 | # Cancel the response before it is completed. |
| 73 | # Poll until the response is no longer queued (started processing) or timeout |
| 74 | loop = asyncio.get_running_loop() |
| 75 | start_time = loop.time() |
| 76 | max_wait_seconds = 5.0 |
| 77 | poll_interval = 0.1 |
| 78 | while loop.time() - start_time < max_wait_seconds: |
| 79 | response = await client.responses.retrieve(response.id) |
| 80 | if response.status != "queued": |
| 81 | # Started processing or completed - try to cancel |
| 82 | break |
| 83 | await asyncio.sleep(poll_interval) |
| 84 | |
| 85 | response = await client.responses.cancel(response.id) |
| 86 | assert response.status == "cancelled" |
| 87 | |
| 88 | # Make sure the response status remains unchanged after some time. |
| 89 | max_retries = 10 |
| 90 | for _ in range(max_retries): |
| 91 | await asyncio.sleep(0.5) |
| 92 | response = await client.responses.retrieve(response.id) |
| 93 | # Verify status is still cancelled |
| 94 | assert response.status == "cancelled" |
| 95 | |
| 96 | |
| 97 | @pytest.mark.asyncio |
| 98 | async def test_cancel_completed(client: openai.AsyncOpenAI): |
| 99 | response = await client.responses.create(input="Hello") |
| 100 | assert response.status == "completed" |
| 101 | |
| 102 | with pytest.raises( |
| 103 | openai.BadRequestError, match="Cannot cancel a synchronous response." |
| 104 | ): |
| 105 | await client.responses.cancel(response.id) |
| 106 | |
| 107 | |
| 108 | @pytest.mark.asyncio |
| 109 | async def test_previous_response_id(client: openai.AsyncOpenAI): |
| 110 | response1 = await client.responses.create( |
| 111 | instructions="You are tested on your ability to retrieve the correct " |
| 112 | "information from the previous response.", |
| 113 | input="Hello, my name is John.", |
| 114 | ) |
| 115 | |
| 116 | response2 = await client.responses.create( |
| 117 | input="Actually, my name is not John. My real name is Mark.", |
| 118 | previous_response_id=response1.id, |
| 119 | ) |
| 120 | |
| 121 | response3 = await client.responses.create( |
| 122 | input="What is my real name again? Answer in one word.", |
| 123 | previous_response_id=response2.id, |
| 124 | ) |
| 125 | print(response3) |
| 126 | assert "Mark" in response3.output[-1].content[0].text |
| 127 | assert "John" not in response3.output[-1].content[0].text |
| 128 | |
| 129 | |
| 130 | @pytest.mark.asyncio |
| 131 | async def test_two_responses_with_same_prev_id(client: openai.AsyncOpenAI): |
| 132 | response1 = await client.responses.create( |
| 133 | instructions="You are tested on your ability to retrieve the correct " |
| 134 | "information from the previous response.", |
| 135 | input="Hello, my name is John.", |
| 136 | ) |
| 137 | |
| 138 | # Both response 2 and 3 use response 1 as the previous response. |
| 139 | response2 = client.responses.create( |
| 140 | input="Actually, my name is not John. My name is Mark.", |
| 141 | previous_response_id=response1.id, |
| 142 | ) |
| 143 | response3 = client.responses.create( |
| 144 | input="What is my name again? Answer in one word.", |
| 145 | previous_response_id=response1.id, |
| 146 | ) |
| 147 | |
| 148 | _ = await response2 |
| 149 | response3_result = await response3 |
| 150 | print(response3_result) |
| 151 | assert "John" in response3_result.output[-1].content[0].text |
| 152 | assert "Mark" not in response3_result.output[-1].content[0].text |