|
| 1 | +# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved. |
| 2 | +# SPDX-License-Identifier: Apache-2.0 |
| 3 | + |
| 4 | + |
| 5 | +from __future__ import annotations |
| 6 | + |
| 7 | +import logging |
| 8 | +import os |
| 9 | +import shutil |
| 10 | +import time |
| 11 | +from typing import Any, Dict |
| 12 | + |
| 13 | +import pytest |
| 14 | +import requests |
| 15 | + |
| 16 | +from tests.conftest import EtcdServer, NatsServer |
| 17 | +from tests.utils.constants import QWEN |
| 18 | +from tests.utils.managed_process import ManagedProcess |
| 19 | +from tests.utils.payloads import check_models_api |
| 20 | + |
| 21 | +logger = logging.getLogger(__name__) |
| 22 | + |
| 23 | +TEST_MODEL = QWEN |
| 24 | + |
| 25 | + |
| 26 | +class DynamoFrontendProcess(ManagedProcess): |
| 27 | + """Process manager for Dynamo frontend""" |
| 28 | + |
| 29 | + def __init__(self, request): |
| 30 | + command = ["python", "-m", "dynamo.frontend", "--router-mode", "round-robin"] |
| 31 | + |
| 32 | + log_dir = f"{request.node.name}_frontend" |
| 33 | + |
| 34 | + # Clean up any existing log directory from previous runs |
| 35 | + try: |
| 36 | + shutil.rmtree(log_dir) |
| 37 | + logger.info(f"Cleaned up existing log directory: {log_dir}") |
| 38 | + except FileNotFoundError: |
| 39 | + # Directory doesn't exist, which is fine |
| 40 | + pass |
| 41 | + |
| 42 | + super().__init__( |
| 43 | + command=command, |
| 44 | + display_output=True, |
| 45 | + terminate_existing=True, |
| 46 | + log_dir=log_dir, |
| 47 | + ) |
| 48 | + |
| 49 | + |
| 50 | +class MockWorkerProcess(ManagedProcess): |
| 51 | + def __init__(self, request, worker_id: str = "mocker-worker"): |
| 52 | + self.worker_id = worker_id |
| 53 | + |
| 54 | + command = [ |
| 55 | + "python3", |
| 56 | + "-m", |
| 57 | + "dynamo.mocker", |
| 58 | + "--model-path", |
| 59 | + TEST_MODEL, |
| 60 | + "--speedup-ratio", |
| 61 | + "100", |
| 62 | + ] |
| 63 | + |
| 64 | + env = os.environ.copy() |
| 65 | + env["DYN_LOG"] = "debug" |
| 66 | + env["DYN_SYSTEM_ENABLED"] = "true" |
| 67 | + env["DYN_SYSTEM_USE_ENDPOINT_HEALTH_STATUS"] = '["generate"]' |
| 68 | + env["DYN_SYSTEM_PORT"] = "8083" |
| 69 | + |
| 70 | + log_dir = f"{request.node.name}_{worker_id}" |
| 71 | + |
| 72 | + try: |
| 73 | + shutil.rmtree(log_dir) |
| 74 | + except FileNotFoundError: |
| 75 | + pass |
| 76 | + |
| 77 | + super().__init__( |
| 78 | + command=command, |
| 79 | + env=env, |
| 80 | + health_check_urls=[ |
| 81 | + ("http://localhost:8000/v1/models", check_models_api), |
| 82 | + ("http://localhost:8083/health", self.is_ready), |
| 83 | + ], |
| 84 | + timeout=300, |
| 85 | + display_output=True, |
| 86 | + terminate_existing=False, |
| 87 | + stragglers=["VLLM::EngineCore"], |
| 88 | + straggler_commands=["-m dynamo.mocker"], |
| 89 | + log_dir=log_dir, |
| 90 | + ) |
| 91 | + |
| 92 | + def is_ready(self, response) -> bool: |
| 93 | + try: |
| 94 | + status = (response.json() or {}).get("status") |
| 95 | + except ValueError: |
| 96 | + logger.warning("%s health response is not valid JSON", self.worker_id) |
| 97 | + return False |
| 98 | + |
| 99 | + is_ready = status == "ready" |
| 100 | + if is_ready: |
| 101 | + logger.info("%s status is ready", self.worker_id) |
| 102 | + else: |
| 103 | + logger.warning("%s status is not ready: %s", self.worker_id, status) |
| 104 | + return is_ready |
| 105 | + |
| 106 | + |
| 107 | +def _send_completion_request( |
| 108 | + payload: Dict[str, Any], |
| 109 | + timeout: int = 180, |
| 110 | +) -> requests.Response: |
| 111 | + """Send a text completion request""" |
| 112 | + |
| 113 | + headers = {"Content-Type": "application/json"} |
| 114 | + print(f"Sending request: {time.time()}") |
| 115 | + |
| 116 | + response = requests.post( |
| 117 | + "http://localhost:8000/v1/completions", |
| 118 | + headers=headers, |
| 119 | + json=payload, |
| 120 | + timeout=timeout, |
| 121 | + ) |
| 122 | + return response |
| 123 | + |
| 124 | + |
| 125 | +@pytest.fixture(scope="module") |
| 126 | +def runtime_services(request): |
| 127 | + """Module-scoped runtime services for this test file.""" |
| 128 | + with NatsServer(request) as nats_process: |
| 129 | + with EtcdServer(request) as etcd_process: |
| 130 | + yield nats_process, etcd_process |
| 131 | + |
| 132 | + |
| 133 | +@pytest.fixture(scope="module") |
| 134 | +def start_services(request, runtime_services): |
| 135 | + """Start frontend and worker processes once for this module's tests.""" |
| 136 | + with DynamoFrontendProcess(request): |
| 137 | + logger.info("Frontend started for tests") |
| 138 | + with MockWorkerProcess(request): |
| 139 | + logger.info("Worker started for tests") |
| 140 | + yield |
| 141 | + |
| 142 | + |
| 143 | +@pytest.mark.usefixtures("start_services") |
| 144 | +@pytest.mark.e2e |
| 145 | +@pytest.mark.model(TEST_MODEL) |
| 146 | +def test_completion_string_prompt() -> None: |
| 147 | + payload: Dict[str, Any] = { |
| 148 | + "model": TEST_MODEL, |
| 149 | + "prompt": "Tell me about Mars", |
| 150 | + "max_tokens": 2000, |
| 151 | + } |
| 152 | + |
| 153 | + response = _send_completion_request(payload) |
| 154 | + |
| 155 | + assert response.status_code == 200, ( |
| 156 | + f"Completion request failed with status " |
| 157 | + f"{response.status_code}: {response.text}" |
| 158 | + ) |
| 159 | + |
| 160 | + |
| 161 | +@pytest.mark.usefixtures("start_services") |
| 162 | +@pytest.mark.e2e |
| 163 | +@pytest.mark.model(TEST_MODEL) |
| 164 | +def test_completion_single_element_array_prompt() -> None: |
| 165 | + payload: Dict[str, Any] = { |
| 166 | + "model": TEST_MODEL, |
| 167 | + "prompt": ["Tell me about Mars"], |
| 168 | + "max_tokens": 2000, |
| 169 | + } |
| 170 | + |
| 171 | + response = _send_completion_request(payload) |
| 172 | + |
| 173 | + assert response.status_code == 200, ( |
| 174 | + f"Completion request failed with status " |
| 175 | + f"{response.status_code}: {response.text}" |
| 176 | + ) |
| 177 | + |
| 178 | + |
| 179 | +@pytest.mark.usefixtures("start_services") |
| 180 | +@pytest.mark.e2e |
| 181 | +@pytest.mark.model(TEST_MODEL) |
| 182 | +def test_completion_multi_element_array_prompt() -> None: |
| 183 | + payload: Dict[str, Any] = { |
| 184 | + "model": TEST_MODEL, |
| 185 | + "prompt": ["Tell me about Mars", "Tell me about Ceres"], |
| 186 | + "max_tokens": 2000, |
| 187 | + } |
| 188 | + |
| 189 | + response = _send_completion_request(payload) |
| 190 | + |
| 191 | + # request should fail because we are sending multiple prompts |
| 192 | + assert ( |
| 193 | + response.status_code == 500 |
| 194 | + ), f"Request should fail with code 500; response:{response.text}" |
0 commit comments