61 lines
1.6 KiB
Python
61 lines
1.6 KiB
Python
import io
|
|
import unittest
|
|
|
|
import requests
|
|
|
|
from sglang.srt.utils import kill_process_tree
|
|
from sglang.test.ci.ci_register import register_cuda_ci
|
|
from sglang.test.test_utils import (
|
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
DEFAULT_URL_FOR_TEST,
|
|
CustomTestCase,
|
|
popen_launch_server,
|
|
)
|
|
|
|
register_cuda_ci(est_time=60, suite="nightly-1-gpu", nightly=True)
|
|
|
|
|
|
class TestRequestLogger(CustomTestCase):
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.stdout = io.StringIO()
|
|
cls.stderr = io.StringIO()
|
|
|
|
cls.process = popen_launch_server(
|
|
"Qwen/Qwen3-0.6B",
|
|
DEFAULT_URL_FOR_TEST,
|
|
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
|
other_args=[
|
|
"--log-requests",
|
|
"--log-requests-level",
|
|
"2",
|
|
"--skip-server-warmup",
|
|
],
|
|
return_stdout_stderr=(cls.stdout, cls.stderr),
|
|
)
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
kill_process_tree(cls.process.pid)
|
|
cls.stdout.close()
|
|
cls.stderr.close()
|
|
|
|
def test_request_logging(self):
|
|
response = requests.post(
|
|
DEFAULT_URL_FOR_TEST + "/generate",
|
|
json={
|
|
"text": "Hello",
|
|
"sampling_params": {"max_new_tokens": 8, "temperature": 0},
|
|
},
|
|
timeout=30,
|
|
)
|
|
self.assertEqual(response.status_code, 200)
|
|
|
|
combined_output = self.stdout.getvalue() + self.stderr.getvalue()
|
|
self.assertIn("Receive:", combined_output)
|
|
self.assertIn("Finish:", combined_output)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|