mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-12-24 13:28:13 +08:00
3cc182236a
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Deploy GitHub Pages / deploy (push) Has been cancelled
434 lines
15 KiB
Python
434 lines
15 KiB
Python
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
||
#
|
||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||
# you may not use this file except in compliance with the License.
|
||
# You may obtain a copy of the License at
|
||
#
|
||
# http://www.apache.org/licenses/LICENSE-2.0
|
||
#
|
||
# Unless required by applicable law or agreed to in writing, software
|
||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||
# See the License for the specific language governing permissions and
|
||
# limitations under the License.
|
||
|
||
import json
|
||
import os
|
||
import signal
|
||
import socket
|
||
import subprocess
|
||
import sys
|
||
import time
|
||
|
||
import pytest
|
||
import requests
|
||
|
||
# Read ports from environment variables; use default values if not set
|
||
FD_API_PORT = int(os.getenv("FD_API_PORT", 8188))
|
||
FD_ENGINE_QUEUE_PORT = int(os.getenv("FD_ENGINE_QUEUE_PORT", 8133))
|
||
FD_METRICS_PORT = int(os.getenv("FD_METRICS_PORT", 8233))
|
||
|
||
# List of ports to clean before and after tests
|
||
PORTS_TO_CLEAN = [
|
||
FD_API_PORT,
|
||
FD_ENGINE_QUEUE_PORT,
|
||
FD_METRICS_PORT,
|
||
FD_API_PORT + 1,
|
||
FD_ENGINE_QUEUE_PORT + 1,
|
||
FD_METRICS_PORT + 1,
|
||
]
|
||
|
||
|
||
def is_port_open(host: str, port: int, timeout=1.0):
|
||
"""
|
||
Check if a TCP port is open on the given host.
|
||
Returns True if connection succeeds, False otherwise.
|
||
"""
|
||
try:
|
||
with socket.create_connection((host, port), timeout):
|
||
return True
|
||
except Exception:
|
||
return False
|
||
|
||
|
||
def kill_process_on_port(port: int):
|
||
"""
|
||
Kill processes that are listening on the given port.
|
||
Uses `lsof` to find process ids and sends SIGKILL.
|
||
"""
|
||
try:
|
||
output = subprocess.check_output(f"lsof -i:{port} -t", shell=True).decode().strip()
|
||
current_pid = os.getpid()
|
||
parent_pid = os.getppid()
|
||
for pid in output.splitlines():
|
||
pid = int(pid)
|
||
if pid in (current_pid, parent_pid):
|
||
print(f"Skip killing current process (pid={pid}) on port {port}")
|
||
continue
|
||
os.kill(pid, signal.SIGKILL)
|
||
print(f"Killed process on port {port}, pid={pid}")
|
||
except subprocess.CalledProcessError:
|
||
pass
|
||
|
||
|
||
def clean_ports():
|
||
"""
|
||
Kill all processes occupying the ports listed in PORTS_TO_CLEAN.
|
||
"""
|
||
for port in PORTS_TO_CLEAN:
|
||
kill_process_on_port(port)
|
||
|
||
|
||
@pytest.fixture(scope="session", autouse=True)
|
||
def setup_and_run_server():
|
||
"""
|
||
Pytest fixture that runs once per test session:
|
||
- Cleans ports before tests
|
||
- Starts the API server as a subprocess
|
||
- Waits for server port to open (up to 30 seconds)
|
||
- Tears down server after all tests finish
|
||
"""
|
||
print("Pre-test port cleanup...")
|
||
clean_ports()
|
||
|
||
base_path = os.getenv("MODEL_PATH")
|
||
if base_path:
|
||
model_path = os.path.join(base_path, "ERNIE-4.5-0.3B-Paddle")
|
||
else:
|
||
model_path = "./ERNIE-4.5-0.3B-Paddle"
|
||
|
||
# prefill实例
|
||
env_prefill = os.environ.copy()
|
||
env_prefill["CUDA_VISIBLE_DEVICES"] = "0"
|
||
env_prefill["INFERENCE_MSG_QUEUE_ID"] = str(FD_API_PORT)
|
||
prefill_log_path = "server.log"
|
||
prefill_cmd = [
|
||
sys.executable,
|
||
"-m",
|
||
"fastdeploy.entrypoints.openai.api_server",
|
||
"--model",
|
||
model_path,
|
||
"--port",
|
||
str(FD_API_PORT),
|
||
"--tensor-parallel-size",
|
||
"1",
|
||
"--engine-worker-queue-port",
|
||
str(FD_ENGINE_QUEUE_PORT),
|
||
"--metrics-port",
|
||
str(FD_METRICS_PORT),
|
||
"--max-model-len",
|
||
"8192",
|
||
"--max-num-seqs",
|
||
"20",
|
||
"--quantization",
|
||
"wint8",
|
||
"--splitwise-role",
|
||
"prefill",
|
||
]
|
||
|
||
# Start subprocess in new process group
|
||
with open(prefill_log_path, "w") as logfile:
|
||
process_prefill = subprocess.Popen(
|
||
prefill_cmd,
|
||
stdout=logfile,
|
||
stderr=subprocess.STDOUT,
|
||
start_new_session=True, # Enables killing full group via os.killpg
|
||
env=env_prefill,
|
||
)
|
||
|
||
# decode实例
|
||
env_decode = os.environ.copy()
|
||
env_decode["CUDA_VISIBLE_DEVICES"] = "1"
|
||
env_decode["INFERENCE_MSG_QUEUE_ID"] = str(FD_API_PORT + 1)
|
||
env_decode["FD_LOG_DIR"] = "decode_log"
|
||
decode_log_path = "decode_server.log"
|
||
decode_cmd = [
|
||
sys.executable,
|
||
"-m",
|
||
"fastdeploy.entrypoints.openai.api_server",
|
||
"--model",
|
||
model_path,
|
||
"--port",
|
||
str(FD_API_PORT + 1),
|
||
"--tensor-parallel-size",
|
||
"1",
|
||
"--engine-worker-queue-port",
|
||
str(FD_ENGINE_QUEUE_PORT + 1),
|
||
"--metrics-port",
|
||
str(FD_METRICS_PORT + 1),
|
||
"--cache-queue-port",
|
||
str(FD_API_PORT + 2),
|
||
"--max-model-len",
|
||
"8192",
|
||
"--max-num-seqs",
|
||
"20",
|
||
"--quantization",
|
||
"wint8",
|
||
"--splitwise-role",
|
||
"decode",
|
||
]
|
||
|
||
# Start subprocess in new process group
|
||
with open(decode_log_path, "w") as logfile:
|
||
process_decode = subprocess.Popen(
|
||
decode_cmd,
|
||
stdout=logfile,
|
||
stderr=subprocess.STDOUT,
|
||
start_new_session=True, # Enables killing full group via os.killpg
|
||
env=env_decode,
|
||
)
|
||
|
||
# Wait up to 300 seconds for API server to be ready
|
||
for _ in range(300):
|
||
if is_port_open("127.0.0.1", FD_API_PORT):
|
||
if is_port_open("127.0.0.1", FD_API_PORT + 1):
|
||
print(f"Prefill server is up on port {FD_API_PORT}")
|
||
print(f"Decode server is up on port {FD_API_PORT + 1}")
|
||
break
|
||
time.sleep(1)
|
||
else:
|
||
print("[TIMEOUT] API server failed to start in 5 minutes. Cleaning up...")
|
||
try:
|
||
os.killpg(process_prefill.pid, signal.SIGTERM)
|
||
os.killpg(process_decode.pid, signal.SIGTERM)
|
||
clean_ports()
|
||
except Exception as e:
|
||
print(f"Failed to kill process group: {e}")
|
||
raise RuntimeError(f"API server did not start on port {FD_API_PORT}")
|
||
|
||
yield # Run tests
|
||
|
||
print("\n===== Post-test server cleanup... =====")
|
||
try:
|
||
os.killpg(process_prefill.pid, signal.SIGTERM)
|
||
os.killpg(process_decode.pid, signal.SIGTERM)
|
||
clean_ports()
|
||
print(f"Prefill server (pid={process_prefill.pid}) terminated")
|
||
print(f"Decode server (pid={process_decode.pid}) terminated")
|
||
except Exception as e:
|
||
print(f"Failed to terminate API server: {e}")
|
||
|
||
|
||
@pytest.fixture(scope="session")
|
||
def api_url(request):
|
||
"""
|
||
Returns the API endpoint URL for chat completions.
|
||
"""
|
||
return f"http://0.0.0.0:{FD_API_PORT}/v1/chat/completions", f"http://0.0.0.0:{FD_API_PORT + 1}/v1/chat/completions"
|
||
|
||
|
||
@pytest.fixture(scope="session")
|
||
def metrics_url(request):
|
||
"""
|
||
Returns the metrics endpoint URL.
|
||
"""
|
||
return f"http://0.0.0.0:{FD_METRICS_PORT}/metrics"
|
||
|
||
|
||
@pytest.fixture
|
||
def headers():
|
||
"""
|
||
Returns common HTTP request headers.
|
||
"""
|
||
return {"Content-Type": "application/json"}
|
||
|
||
|
||
def send_request(url, payload, timeout=600):
|
||
"""
|
||
发送请求到指定的URL，并返回响应结果。
|
||
"""
|
||
headers = {
|
||
"Content-Type": "application/json",
|
||
}
|
||
|
||
try:
|
||
res = requests.post(url, headers=headers, json=payload, timeout=timeout)
|
||
print("🟢 接收响应中...\n")
|
||
return res
|
||
except requests.exceptions.Timeout:
|
||
print(f"❌ 请求超时（超过 {timeout} 秒）")
|
||
return None
|
||
except requests.exceptions.RequestException as e:
|
||
print(f"❌ 请求失败：{e}")
|
||
return None
|
||
|
||
|
||
def get_stream_chunks(response):
|
||
"""解析流式返回，生成chunk List[dict]"""
|
||
chunks = []
|
||
|
||
if response.status_code == 200:
|
||
for line in response.iter_lines(decode_unicode=True):
|
||
if line:
|
||
if line.startswith("data: "):
|
||
line = line[len("data: ") :]
|
||
|
||
if line.strip() == "[DONE]":
|
||
break
|
||
|
||
try:
|
||
chunk = json.loads(line)
|
||
chunks.append(chunk)
|
||
except Exception as e:
|
||
print(f"解析失败: {e}, 行内容: {line}")
|
||
else:
|
||
print(f"请求失败，状态码: {response.status_code}")
|
||
print("返回内容：", response.text)
|
||
|
||
return chunks
|
||
|
||
|
||
def test_chat_usage_stream(api_url):
|
||
"""测试流式chat usage"""
|
||
payload = {
|
||
"model": "default",
|
||
"temperature": 0,
|
||
"top_p": 0,
|
||
"seed": 33,
|
||
"messages": [
|
||
{"role": "system", "content": "You are a helpful assistant."},
|
||
{"role": "user", "content": "牛顿的三大运动定律是什么？"},
|
||
],
|
||
"max_tokens": 50,
|
||
"stream": True,
|
||
"stream_options": {"include_usage": True, "continuous_usage_stats": True},
|
||
"metadata": {"min_tokens": 10},
|
||
}
|
||
p_url, d_url = api_url
|
||
|
||
response = send_request(url=p_url, payload=payload)
|
||
chunks = get_stream_chunks(response)
|
||
result = "".join([x["choices"][0]["delta"]["content"] for x in chunks[:-1]])
|
||
print("Prefill Response:", result)
|
||
assert result != "", "结果为空"
|
||
usage = chunks[-1]["usage"]
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
response = send_request(url=d_url, payload=payload)
|
||
chunks = get_stream_chunks(response)
|
||
result = "".join([x["choices"][0]["delta"]["content"] for x in chunks[:-1]])
|
||
print("Decode Response:", result)
|
||
assert result != "", "结果为空"
|
||
# for idx, chunk in enumerate(chunks):
|
||
# print(f"\nchunk[{idx}]:\n{json.dumps(chunk, indent=2, ensure_ascii=False)}")
|
||
usage = chunks[-1]["usage"]
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
|
||
def test_chat_usage_non_stream(api_url):
|
||
"""测试非流式chat usage"""
|
||
payload = {
|
||
"model": "default",
|
||
"temperature": 0,
|
||
"top_p": 0,
|
||
"seed": 33,
|
||
"messages": [
|
||
{"role": "system", "content": "You are a helpful assistant."},
|
||
{"role": "user", "content": "牛顿的三大运动定律是什么？"},
|
||
],
|
||
"max_tokens": 50,
|
||
"stream": False,
|
||
"metadata": {"min_tokens": 10},
|
||
}
|
||
p_url, d_url = api_url
|
||
|
||
response = send_request(url=p_url, payload=payload).json()
|
||
usage = response["usage"]
|
||
result = response["choices"][0]["message"]["content"]
|
||
assert result != "", "结果为空"
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
response = send_request(url=d_url, payload=payload).json()
|
||
usage = response["usage"]
|
||
result = response["choices"][0]["message"]["content"]
|
||
assert result != "", "结果为空"
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
|
||
def test_non_chat_usage_stream(api_url):
|
||
"""测试流式非chat usage"""
|
||
payload = {
|
||
"model": "default",
|
||
"temperature": 0,
|
||
"top_p": 0,
|
||
"seed": 33,
|
||
"prompt": "牛顿的三大运动定律是什么？",
|
||
"max_tokens": 50,
|
||
"stream": True,
|
||
"stream_options": {"include_usage": True, "continuous_usage_stats": True},
|
||
"metadata": {"min_tokens": 10},
|
||
}
|
||
p_url, d_url = api_url
|
||
p_url = p_url.replace("chat/completions", "completions")
|
||
d_url = d_url.replace("chat/completions", "completions")
|
||
|
||
response = send_request(url=p_url, payload=payload)
|
||
chunks = get_stream_chunks(response)
|
||
result = "".join([x["choices"][0]["text"] for x in chunks[:-1]])
|
||
# print("Prefill Response:", result)
|
||
assert result != "", "结果为空"
|
||
usage = chunks[-1]["usage"]
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
response = send_request(url=d_url, payload=payload)
|
||
chunks = get_stream_chunks(response)
|
||
result = "".join([x["choices"][0]["text"] for x in chunks[:-1]])
|
||
# print("Decode Response:", result)
|
||
assert result != "", "结果为空"
|
||
usage = chunks[-1]["usage"]
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
|
||
def test_non_chat_usage_non_stream(api_url):
|
||
"""测试非流式非chat usage"""
|
||
payload = {
|
||
"model": "default",
|
||
"temperature": 0,
|
||
"top_p": 0,
|
||
"seed": 33,
|
||
"prompt": "牛顿的三大运动定律是什么？",
|
||
"max_tokens": 50,
|
||
"stream": False,
|
||
"metadata": {"min_tokens": 10},
|
||
}
|
||
p_url, d_url = api_url
|
||
p_url = p_url.replace("chat/completions", "completions")
|
||
d_url = d_url.replace("chat/completions", "completions")
|
||
|
||
response = send_request(url=p_url, payload=payload).json()
|
||
usage = response["usage"]
|
||
result = response["choices"][0]["text"]
|
||
# print("Prefill Response:", result)
|
||
assert result != "", "结果为空"
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|
||
|
||
response = send_request(url=d_url, payload=payload).json()
|
||
usage = response["usage"]
|
||
result = response["choices"][0]["text"]
|
||
assert result != "", "结果为空"
|
||
total_tokens = usage["completion_tokens"] + usage["prompt_tokens"]
|
||
assert payload["max_tokens"] >= usage["completion_tokens"], "completion_tokens大于max_tokens"
|
||
assert payload["metadata"]["min_tokens"] <= usage["completion_tokens"], "completion_tokens小于min_tokens"
|
||
assert usage["total_tokens"] == total_tokens, "total_tokens不等于prompt_tokens + completion_tokens"
|