[router] Add gRPC E2E test suite (#11790)
This commit is contained in:
Keyang Ru
204
sgl-router/py_test/e2e_grpc/features/test_enable_thinking.py
Normal file
204
sgl-router/py_test/e2e_grpc/features/test_enable_thinking.py
Normal file
@@ -0,0 +1,204 @@
|
||||
"""
|
||||
Usage:
|
||||
python3 -m unittest openai_server.features.test_enable_thinking.TestEnableThinking.test_chat_completion_with_reasoning
|
||||
python3 -m unittest openai_server.features.test_enable_thinking.TestEnableThinking.test_chat_completion_without_reasoning
|
||||
python3 -m unittest openai_server.features.test_enable_thinking.TestEnableThinking.test_stream_chat_completion_with_reasoning
|
||||
python3 -m unittest openai_server.features.test_enable_thinking.TestEnableThinking.test_stream_chat_completion_without_reasoning
|
||||
"""
|
||||
|
||||
import asyncio
|
||||
import json
|
||||
import os
|
||||
import sys
|
||||
import time
|
||||
import unittest
|
||||
|
||||
# CHANGE: Import router launcher instead of server launcher
|
||||
from pathlib import Path
|
||||
|
||||
import openai
|
||||
import requests
|
||||
|
||||
_TEST_DIR = Path(__file__).parent
|
||||
sys.path.insert(0, str(_TEST_DIR.parent))
|
||||
from fixtures import popen_launch_workers_and_router
|
||||
from util import (
|
||||
DEFAULT_ENABLE_THINKING_MODEL_PATH,
|
||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
get_tokenizer,
|
||||
kill_process_tree,
|
||||
)
|
||||
|
||||
|
||||
class TestEnableThinking(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
# CHANGE: Launch gRPC router with integrated workers (single command)
|
||||
cls.model = DEFAULT_ENABLE_THINKING_MODEL_PATH
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.api_key = "sk-1234"
|
||||
cls.cluster = popen_launch_workers_and_router(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=120,
|
||||
api_key=cls.api_key,
|
||||
router_args=[
|
||||
"--reasoning-parser",
|
||||
"qwen3",
|
||||
],
|
||||
num_workers=1,
|
||||
tp_size=4,
|
||||
)
|
||||
cls.additional_chat_kwargs = {}
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
# Cleanup router and workers
|
||||
kill_process_tree(cls.cluster["router"].pid)
|
||||
for worker in cls.cluster.get("workers", []):
|
||||
kill_process_tree(worker.pid)
|
||||
|
||||
def test_chat_completion_with_reasoning(self):
|
||||
# Test non-streaming with "enable_thinking": True, reasoning_content should not be empty
|
||||
client = requests.post(
|
||||
f"{self.base_url}/v1/chat/completions",
|
||||
headers={"Authorization": f"Bearer {self.api_key}"},
|
||||
json={
|
||||
"model": self.model,
|
||||
"messages": [{"role": "user", "content": "Hello"}],
|
||||
"temperature": 0,
|
||||
"separate_reasoning": True,
|
||||
"chat_template_kwargs": {"enable_thinking": True},
|
||||
**self.additional_chat_kwargs,
|
||||
},
|
||||
)
|
||||
|
||||
self.assertEqual(client.status_code, 200, f"Failed with: {client.text}")
|
||||
data = client.json()
|
||||
|
||||
self.assertIn("choices", data)
|
||||
self.assertTrue(len(data["choices"]) > 0)
|
||||
self.assertIn("message", data["choices"][0])
|
||||
self.assertIn("reasoning_content", data["choices"][0]["message"])
|
||||
self.assertIsNotNone(data["choices"][0]["message"]["reasoning_content"])
|
||||
|
||||
def test_chat_completion_without_reasoning(self):
|
||||
# Test non-streaming with "enable_thinking": False, reasoning_content should be empty
|
||||
client = requests.post(
|
||||
f"{self.base_url}/v1/chat/completions",
|
||||
headers={"Authorization": f"Bearer {self.api_key}"},
|
||||
json={
|
||||
"model": self.model,
|
||||
"messages": [{"role": "user", "content": "Hello"}],
|
||||
"temperature": 0,
|
||||
"separate_reasoning": True,
|
||||
"chat_template_kwargs": {"enable_thinking": False},
|
||||
**self.additional_chat_kwargs,
|
||||
},
|
||||
)
|
||||
|
||||
self.assertEqual(client.status_code, 200, f"Failed with: {client.text}")
|
||||
data = client.json()
|
||||
|
||||
self.assertIn("choices", data)
|
||||
self.assertTrue(len(data["choices"]) > 0)
|
||||
self.assertIn("message", data["choices"][0])
|
||||
|
||||
if "reasoning_content" in data["choices"][0]["message"]:
|
||||
self.assertIsNone(data["choices"][0]["message"]["reasoning_content"])
|
||||
|
||||
def test_stream_chat_completion_with_reasoning(self):
|
||||
# Test streaming with "enable_thinking": True, reasoning_content should not be empty
|
||||
response = requests.post(
|
||||
f"{self.base_url}/v1/chat/completions",
|
||||
headers={"Authorization": f"Bearer {self.api_key}"},
|
||||
json={
|
||||
"model": self.model,
|
||||
"messages": [{"role": "user", "content": "Hello"}],
|
||||
"temperature": 0,
|
||||
"separate_reasoning": True,
|
||||
"stream": True,
|
||||
"chat_template_kwargs": {"enable_thinking": True},
|
||||
**self.additional_chat_kwargs,
|
||||
},
|
||||
stream=True,
|
||||
)
|
||||
|
||||
self.assertEqual(response.status_code, 200, f"Failed with: {response.text}")
|
||||
|
||||
has_reasoning = False
|
||||
has_content = False
|
||||
|
||||
print("\n=== Stream With Reasoning ===")
|
||||
for line in response.iter_lines():
|
||||
if line:
|
||||
line = line.decode("utf-8")
|
||||
if line.startswith("data:") and not line.startswith("data: [DONE]"):
|
||||
data = json.loads(line[6:])
|
||||
if "choices" in data and len(data["choices"]) > 0:
|
||||
delta = data["choices"][0].get("delta", {})
|
||||
|
||||
if "reasoning_content" in delta and delta["reasoning_content"]:
|
||||
has_reasoning = True
|
||||
|
||||
if "content" in delta and delta["content"]:
|
||||
has_content = True
|
||||
|
||||
self.assertTrue(
|
||||
has_reasoning,
|
||||
"The reasoning content is not included in the stream response",
|
||||
)
|
||||
self.assertTrue(
|
||||
has_content, "The stream response does not contain normal content"
|
||||
)
|
||||
|
||||
def test_stream_chat_completion_without_reasoning(self):
|
||||
# Test streaming with "enable_thinking": False, reasoning_content should be empty
|
||||
response = requests.post(
|
||||
f"{self.base_url}/v1/chat/completions",
|
||||
headers={"Authorization": f"Bearer {self.api_key}"},
|
||||
json={
|
||||
"model": self.model,
|
||||
"messages": [{"role": "user", "content": "Hello"}],
|
||||
"temperature": 0,
|
||||
"separate_reasoning": True,
|
||||
"stream": True,
|
||||
"chat_template_kwargs": {"enable_thinking": False},
|
||||
**self.additional_chat_kwargs,
|
||||
},
|
||||
stream=True,
|
||||
)
|
||||
|
||||
self.assertEqual(response.status_code, 200, f"Failed with: {response.text}")
|
||||
|
||||
has_reasoning = False
|
||||
has_content = False
|
||||
|
||||
print("\n=== Stream Without Reasoning ===")
|
||||
for line in response.iter_lines():
|
||||
if line:
|
||||
line = line.decode("utf-8")
|
||||
if line.startswith("data:") and not line.startswith("data: [DONE]"):
|
||||
data = json.loads(line[6:])
|
||||
if "choices" in data and len(data["choices"]) > 0:
|
||||
delta = data["choices"][0].get("delta", {})
|
||||
|
||||
if "reasoning_content" in delta and delta["reasoning_content"]:
|
||||
has_reasoning = True
|
||||
|
||||
if "content" in delta and delta["content"]:
|
||||
has_content = True
|
||||
|
||||
self.assertFalse(
|
||||
has_reasoning,
|
||||
"The reasoning content should not be included in the stream response",
|
||||
)
|
||||
self.assertTrue(
|
||||
has_content, "The stream response does not contain normal content"
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
198
sgl-router/py_test/e2e_grpc/features/test_reasoning_content.py
Normal file
198
sgl-router/py_test/e2e_grpc/features/test_reasoning_content.py
Normal file
@@ -0,0 +1,198 @@
|
||||
"""
|
||||
Usage:
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentAPI.test_streaming_separate_reasoning_false
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentAPI.test_streaming_separate_reasoning_true
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentAPI.test_streaming_separate_reasoning_true_stream_reasoning_false
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentAPI.test_nonstreaming_separate_reasoning_false
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentAPI.test_nonstreaming_separate_reasoning_true
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentStartup.test_nonstreaming
|
||||
python3 -m unittest openai_server.features.test_reasoning_content.TestReasoningContentStartup.test_streaming
|
||||
"""
|
||||
|
||||
import json
|
||||
|
||||
# CHANGE: Import router launcher instead of server launcher
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
import openai
|
||||
import requests
|
||||
|
||||
_TEST_DIR = Path(__file__).parent
|
||||
sys.path.insert(0, str(_TEST_DIR.parent))
|
||||
from fixtures import popen_launch_workers_and_router
|
||||
from util import (
|
||||
DEFAULT_REASONING_MODEL_PATH,
|
||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
kill_process_tree,
|
||||
)
|
||||
|
||||
|
||||
class TestReasoningContentAPI(CustomTestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
# CHANGE: Launch gRPC router with integrated workers (single command)
|
||||
cls.model = DEFAULT_REASONING_MODEL_PATH
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
cls.api_key = "sk-1234"
|
||||
cls.cluster = popen_launch_workers_and_router(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
api_key=cls.api_key,
|
||||
router_args=[
|
||||
"--reasoning-parser",
|
||||
"deepseek_r1",
|
||||
],
|
||||
num_workers=1,
|
||||
tp_size=2,
|
||||
)
|
||||
cls.base_url += "/v1"
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
# Cleanup router and workers
|
||||
kill_process_tree(cls.cluster["router"].pid)
|
||||
for worker in cls.cluster.get("workers", []):
|
||||
kill_process_tree(worker.pid)
|
||||
|
||||
def test_streaming_separate_reasoning_false(self):
|
||||
# Test streaming with separate_reasoning=False, reasoning_content should be empty
|
||||
client = openai.Client(api_key=self.api_key, base_url=self.base_url)
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": "What is 1+3?",
|
||||
}
|
||||
],
|
||||
"max_tokens": 100,
|
||||
"stream": True,
|
||||
"extra_body": {"separate_reasoning": False},
|
||||
}
|
||||
response = client.chat.completions.create(**payload)
|
||||
|
||||
reasoning_content = ""
|
||||
content = ""
|
||||
for chunk in response:
|
||||
if chunk.choices[0].delta.content:
|
||||
content += chunk.choices[0].delta.content
|
||||
elif chunk.choices[0].delta.reasoning_content:
|
||||
reasoning_content += chunk.choices[0].delta.reasoning_content
|
||||
|
||||
assert len(reasoning_content) == 0
|
||||
assert len(content) > 0
|
||||
|
||||
def test_streaming_separate_reasoning_true(self):
|
||||
# Test streaming with separate_reasoning=True, reasoning_content should not be empty
|
||||
client = openai.Client(api_key=self.api_key, base_url=self.base_url)
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": "What is 1+3?",
|
||||
}
|
||||
],
|
||||
"max_tokens": 100,
|
||||
"stream": True,
|
||||
"extra_body": {"separate_reasoning": True},
|
||||
}
|
||||
response = client.chat.completions.create(**payload)
|
||||
|
||||
reasoning_content = ""
|
||||
content = ""
|
||||
for chunk in response:
|
||||
if chunk.choices[0].delta.content:
|
||||
content += chunk.choices[0].delta.content
|
||||
elif chunk.choices[0].delta.reasoning_content:
|
||||
reasoning_content += chunk.choices[0].delta.reasoning_content
|
||||
|
||||
assert len(reasoning_content) > 0
|
||||
assert len(content) > 0
|
||||
|
||||
def test_streaming_separate_reasoning_true_stream_reasoning_false(self):
|
||||
# Test streaming with separate_reasoning=True, reasoning_content should not be empty
|
||||
client = openai.Client(api_key=self.api_key, base_url=self.base_url)
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": "What is 1+3?",
|
||||
}
|
||||
],
|
||||
"max_tokens": 100,
|
||||
"stream": True,
|
||||
"extra_body": {"separate_reasoning": True, "stream_reasoning": False},
|
||||
}
|
||||
response = client.chat.completions.create(**payload)
|
||||
|
||||
reasoning_content = ""
|
||||
content = ""
|
||||
first_chunk = False
|
||||
for chunk in response:
|
||||
if chunk.choices[0].delta.reasoning_content:
|
||||
reasoning_content = chunk.choices[0].delta.reasoning_content
|
||||
first_chunk = True
|
||||
if chunk.choices[0].delta.content:
|
||||
content += chunk.choices[0].delta.content
|
||||
if not first_chunk:
|
||||
reasoning_content = chunk.choices[0].delta.reasoning_content
|
||||
first_chunk = True
|
||||
if not first_chunk:
|
||||
assert (
|
||||
not chunk.choices[0].delta.reasoning_content
|
||||
or len(chunk.choices[0].delta.reasoning_content) == 0
|
||||
)
|
||||
assert len(reasoning_content) > 0
|
||||
assert len(content) > 0
|
||||
|
||||
def test_nonstreaming_separate_reasoning_false(self):
|
||||
# Test non-streaming with separate_reasoning=False, reasoning_content should be empty
|
||||
client = openai.Client(api_key=self.api_key, base_url=self.base_url)
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": "What is 1+3?",
|
||||
}
|
||||
],
|
||||
"max_tokens": 100,
|
||||
"extra_body": {"separate_reasoning": False},
|
||||
}
|
||||
response = client.chat.completions.create(**payload)
|
||||
|
||||
assert (
|
||||
not response.choices[0].message.reasoning_content
|
||||
or len(response.choices[0].message.reasoning_content) == 0
|
||||
)
|
||||
assert len(response.choices[0].message.content) > 0
|
||||
|
||||
def test_nonstreaming_separate_reasoning_true(self):
|
||||
# Test non-streaming with separate_reasoning=True, reasoning_content should not be empty
|
||||
client = openai.Client(api_key=self.api_key, base_url=self.base_url)
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": "What is 1+3?",
|
||||
}
|
||||
],
|
||||
"max_tokens": 100,
|
||||
"extra_body": {"separate_reasoning": True},
|
||||
}
|
||||
response = client.chat.completions.create(**payload)
|
||||
|
||||
assert len(response.choices[0].message.reasoning_content) > 0
|
||||
assert len(response.choices[0].message.content) > 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user