313 lines
10 KiB
Python
313 lines
10 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
系统配置优化 - 调整系统参数以提高并发性能
|
||
"""
|
||
|
||
import os
|
||
import asyncio
|
||
import multiprocessing
|
||
import threading
|
||
import time
|
||
import resource
|
||
import logging
|
||
from typing import Dict, Any, Optional
|
||
from concurrent.futures import ThreadPoolExecutor
|
||
|
||
# 配置日志
|
||
logger = logging.getLogger('app')
|
||
|
||
|
||
class SystemOptimizer:
|
||
"""系统优化器
|
||
|
||
调整系统参数以提高并发性能
|
||
"""
|
||
|
||
def __init__(self):
|
||
self.original_settings = {}
|
||
self.optimized = False
|
||
|
||
def optimize_system_settings(self):
|
||
"""优化系统设置"""
|
||
if self.optimized:
|
||
return
|
||
|
||
# 保存原始设置
|
||
self._backup_original_settings()
|
||
|
||
# 1. 优化文件描述符限制
|
||
try:
|
||
soft, hard = resource.getrlimit(resource.RLIMIT_NOFILE)
|
||
new_limit = min(65536, hard) # 增加到65536或硬件限制
|
||
resource.setrlimit(resource.RLIMIT_NOFILE, (new_limit, hard))
|
||
self.original_settings['RLIMIT_NOFILE'] = (soft, hard)
|
||
logger.info(f"文件描述符限制从 {soft} 增加到 {new_limit}")
|
||
except (ValueError, OSError) as e:
|
||
logger.error(f"无法设置文件描述符限制: {e}")
|
||
|
||
# 2. 优化线程栈大小
|
||
try:
|
||
soft, hard = resource.getrlimit(resource.RLIMIT_STACK)
|
||
new_stack = min(8 * 1024 * 1024, hard) # 8MB栈大小
|
||
resource.setrlimit(resource.RLIMIT_STACK, (new_stack, hard))
|
||
self.original_settings['RLIMIT_STACK'] = (soft, hard)
|
||
logger.info(f"线程栈大小设置为 {new_stack // (1024*1024)}MB")
|
||
except (ValueError, OSError) as e:
|
||
logger.error(f"无法设置线程栈大小: {e}")
|
||
|
||
# 3. 环境变量优化
|
||
env_vars = {
|
||
# Python优化
|
||
'PYTHONUNBUFFERED': '1', # 禁用输出缓冲
|
||
'PYTHONDONTWRITEBYTECODE': '1', # 不写.pyc文件
|
||
|
||
# Tokenizer优化 - 允许适度的并行度
|
||
'TOKENIZERS_PARALLELISM': 'true', # 改为true以提高并发
|
||
'TOKENIZERS_FAST': '1', # 启用快速tokenizer
|
||
|
||
# OpenMP优化
|
||
'OMP_NUM_THREADS': str(min(8, multiprocessing.cpu_count())), # 限制OpenMP线程
|
||
'OMP_WAIT_POLICY': 'PASSIVE', # 被动等待策略
|
||
|
||
# 内存优化
|
||
'MALLOC_TRIM_THRESHOLD_': '100000', # 内存整理阈值
|
||
|
||
# 网络优化
|
||
'TCP_NODELAY': '1', # 禁用Nagle算法
|
||
|
||
# Hugging Face优化
|
||
'TRANSFORMERS_CACHE': '/tmp/transformers_cache', # 使用tmpfs加速缓存
|
||
'HF_OFFLINE': '0', # 在线模式
|
||
|
||
# CUDA优化（如果使用GPU）
|
||
'CUDA_LAUNCH_BLOCKING': '0', # 异步CUDA启动
|
||
|
||
# asyncio优化
|
||
'UVLOOP_ENABLED': '1', # 启用uvloop（如果可用）
|
||
}
|
||
|
||
for key, value in env_vars.items():
|
||
if key not in os.environ:
|
||
os.environ[key] = value
|
||
logger.info(f"设置环境变量: {key}={value}")
|
||
|
||
self.optimized = True
|
||
logger.info("系统优化完成")
|
||
|
||
def _backup_original_settings(self):
|
||
"""备份原始设置"""
|
||
try:
|
||
self.original_settings['RLIMIT_NOFILE'] = resource.getrlimit(resource.RLIMIT_NOFILE)
|
||
self.original_settings['RLIMIT_STACK'] = resource.getrlimit(resource.RLIMIT_STACK)
|
||
except:
|
||
pass
|
||
|
||
# 备份重要环境变量
|
||
env_keys = [
|
||
'TOKENIZERS_PARALLELISM',
|
||
'PYTHONUNBUFFERED',
|
||
'PYTHONDONTWRITEBYTECODE',
|
||
'OMP_NUM_THREADS'
|
||
]
|
||
|
||
for key in env_keys:
|
||
if key in os.environ:
|
||
self.original_settings[key] = os.environ[key]
|
||
|
||
def restore_original_settings(self):
|
||
"""恢复原始设置"""
|
||
if not self.original_settings:
|
||
return
|
||
|
||
logger.info("恢复原始系统设置...")
|
||
|
||
# 恢复资源限制
|
||
if 'RLIMIT_NOFILE' in self.original_settings:
|
||
try:
|
||
resource.setrlimit(resource.RLIMIT_NOFILE, self.original_settings['RLIMIT_NOFILE'])
|
||
logger.info(f"恢复文件描述符限制")
|
||
except:
|
||
pass
|
||
|
||
if 'RLIMIT_STACK' in self.original_settings:
|
||
try:
|
||
resource.setrlimit(resource.RLIMIT_STACK, self.original_settings['RLIMIT_STACK'])
|
||
logger.info(f"恢复线程栈大小")
|
||
except:
|
||
pass
|
||
|
||
# 恢复环境变量
|
||
for key, value in self.original_settings.items():
|
||
if key.startswith('TOKENIZERS_') or key in ['PYTHONUNBUFFERED', 'PYTHONDONTWRITEBYTECODE']:
|
||
if key in os.environ:
|
||
del os.environ[key]
|
||
logger.info(f"移除环境变量: {key}")
|
||
elif key in ['OMP_NUM_THREADS'] and value is not None:
|
||
os.environ[key] = value
|
||
logger.info(f"恢复环境变量: {key}={value}")
|
||
|
||
self.optimized = False
|
||
logger.info("系统设置已恢复")
|
||
|
||
|
||
class AsyncioOptimizer:
|
||
"""asyncio优化器"""
|
||
|
||
@staticmethod
|
||
def setup_event_loop_policy():
|
||
"""设置优化的事件循环策略"""
|
||
try:
|
||
# 尝试使用uvloop（如果可用）
|
||
import uvloop
|
||
asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
|
||
logger.info("使用uvloop事件循环策略")
|
||
except ImportError:
|
||
logger.info("使用默认事件循环策略")
|
||
|
||
# 设置线程池大小
|
||
cpu_count = multiprocessing.cpu_count()
|
||
thread_pool_size = min(32, cpu_count * 4) # 每CPU核心4个线程，最多32个
|
||
|
||
# 注意：不能在这里设置默认线程池执行器，因为还没有运行事件循环
|
||
# 这个设置会在应用启动时进行
|
||
logger.info(f"建议线程池大小: {thread_pool_size}")
|
||
|
||
@staticmethod
|
||
def optimize_gunicorn_settings() -> Dict[str, Any]:
|
||
"""获取优化的Gunicorn设置"""
|
||
cpu_count = multiprocessing.cpu_count()
|
||
|
||
return {
|
||
# Worker配置
|
||
'workers': min(8, cpu_count + 1), # 工作进程数
|
||
'worker_class': 'uvicorn.workers.UvicornWorker', # 使用Uvicorn worker
|
||
'worker_connections': 2000, # 每个worker的连接数
|
||
'max_requests': 5000, # 最大请求数后重启worker
|
||
'max_requests_jitter': 500, # 随机抖动
|
||
'preload_app': True, # 预加载应用
|
||
|
||
# 超时设置
|
||
'timeout': 120, # 工作超时
|
||
'keepalive': 5, # Keep-Alive超时
|
||
'graceful_timeout': 30, # 优雅关闭超时
|
||
|
||
# 性能优化
|
||
'worker_tmp_dir': '/dev/shm', # 使用内存文件系统
|
||
|
||
# 日志设置
|
||
'accesslog': '-', # 标准输出
|
||
'errorlog': '-', # 标准错误输出
|
||
'loglevel': 'info',
|
||
}
|
||
|
||
|
||
def setup_system_optimizations():
|
||
"""设置系统优化"""
|
||
# 1. 系统级优化
|
||
system_optimizer = SystemOptimizer()
|
||
system_optimizer.optimize_system_settings()
|
||
|
||
# 2. asyncio优化
|
||
asyncio_optimizer = AsyncioOptimizer()
|
||
asyncio_optimizer.setup_event_loop_policy()
|
||
|
||
return system_optimizer
|
||
|
||
|
||
def create_performance_monitor() -> Dict[str, Any]:
|
||
"""创建性能监控配置"""
|
||
return {
|
||
'monitor_interval': 60, # 监控间隔（秒）
|
||
'metrics': {
|
||
'memory_usage': True,
|
||
'cpu_usage': True,
|
||
'disk_io': True,
|
||
'network_io': True,
|
||
'active_connections': True,
|
||
'request_latency': True,
|
||
'cache_hit_rate': True,
|
||
'error_rate': True,
|
||
},
|
||
'alerts': {
|
||
'memory_threshold': 0.9, # 90%内存使用率告警
|
||
'cpu_threshold': 0.8, # 80%CPU使用率告警
|
||
'disk_threshold': 0.9, # 90%磁盘使用率告警
|
||
'error_threshold': 0.05, # 5%错误率告警
|
||
'latency_threshold': 5.0, # 5秒延迟告警
|
||
}
|
||
}
|
||
|
||
|
||
def get_optimized_worker_config() -> Dict[str, Any]:
|
||
"""获取优化的worker配置"""
|
||
cpu_count = multiprocessing.cpu_count()
|
||
memory_gb = os.sysconf('SC_PAGE_SIZE') * os.sysconf('SC_PHYS_PAGES') / (1024.0 ** 3)
|
||
|
||
# 基于资源限制的配置
|
||
max_workers = min(
|
||
16, # 最大worker数
|
||
max(2, cpu_count), # 至少2个worker，最多CPU核心数
|
||
int(memory_gb / 2) # 基于内存的worker限制（每worker 2GB）
|
||
)
|
||
|
||
return {
|
||
'max_workers': max_workers,
|
||
'worker_connections': 1000, # 每个worker的连接数
|
||
'connection_pool_size': 100, # 连接池大小
|
||
'buffer_size': 8192, # 缓冲区大小
|
||
'timeout': 120, # 超时时间
|
||
'keepalive_timeout': 30, # Keep-Alive超时
|
||
}
|
||
|
||
|
||
# 预定义的优化配置
|
||
OPTIMIZATION_CONFIGS = {
|
||
'low_memory': {
|
||
'max_workers': 2,
|
||
'worker_connections': 500,
|
||
'buffer_size': 4096,
|
||
'cache_size': 500,
|
||
},
|
||
'balanced': {
|
||
'max_workers': 4,
|
||
'worker_connections': 1000,
|
||
'buffer_size': 8192,
|
||
'cache_size': 1000,
|
||
},
|
||
'high_performance': {
|
||
'max_workers': 8,
|
||
'worker_connections': 2000,
|
||
'buffer_size': 16384,
|
||
'cache_size': 2000,
|
||
}
|
||
}
|
||
|
||
|
||
def apply_optimization_profile(profile_name: str) -> Dict[str, Any]:
|
||
"""应用优化配置文件"""
|
||
if profile_name not in OPTIMIZATION_CONFIGS:
|
||
raise ValueError(f"未知的优化配置: {profile_name}")
|
||
|
||
config = OPTIMIZATION_CONFIGS[profile_name].copy()
|
||
|
||
# 添加系统特定配置
|
||
config.update({
|
||
'profile_name': profile_name,
|
||
'applied_at': time.time(),
|
||
'cpu_count': multiprocessing.cpu_count(),
|
||
'memory_gb': os.sysconf('SC_PAGE_SIZE') * os.sysconf('SC_PHYS_PAGES') / (1024.0 ** 3)
|
||
})
|
||
|
||
return config
|
||
|
||
|
||
# 全局系统优化器实例
|
||
_global_system_optimizer: Optional[SystemOptimizer] = None
|
||
|
||
|
||
def get_global_system_optimizer() -> SystemOptimizer:
|
||
"""获取全局系统优化器"""
|
||
global _global_system_optimizer
|
||
if _global_system_optimizer is None:
|
||
_global_system_optimizer = SystemOptimizer()
|
||
return _global_system_optimizer |