dekun
481 lines
15 KiB
Bash
481 lines
15 KiB
Bash
#!/usr/bin/env bash
|
||
# =============================================================================
|
||
# Trading Studio 一键部署脚本
|
||
# 安装路径: /opt/Trading_Studio
|
||
# 运行用户: root
|
||
# 功能: 系统依赖 → 代码拉取 → Python 虚拟环境 → PyTorch CUDA → PM2 常驻
|
||
#
|
||
# 用法:
|
||
# sudo bash deploy.sh # 首次完整部署并 PM2 启动
|
||
# sudo bash deploy.sh update # 拉取最新代码、更新依赖、重启 PM2
|
||
# sudo bash deploy.sh restart # 仅重启 PM2 进程
|
||
# sudo bash deploy.sh stop # 停止 PM2 进程
|
||
# sudo bash deploy.sh status # 查看 PM2 与 GPU 状态
|
||
# sudo bash deploy.sh logs # 查看 PM2 最近日志
|
||
#
|
||
# 环境变量:
|
||
# USE_CN_MIRROR=1 使用清华 PyPI / PyTorch 镜像（国内服务器推荐，默认开启）
|
||
# USE_CN_MIRROR=0 使用官方 PyTorch 源
|
||
# PIP_TIMEOUT=600 pip 单次下载超时秒数（默认 600）
|
||
# SKIP_PYTORCH=1 跳过 PyTorch 安装（已手动装好时使用）
|
||
# =============================================================================
|
||
|
||
set -euo pipefail
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 可配置常量
|
||
# ---------------------------------------------------------------------------
|
||
INSTALL_DIR="/opt/Trading_Studio"
|
||
GIT_REPO="https://git.bz121.com/dekun/Trading_Studio.git"
|
||
GIT_BRANCH="main"
|
||
PM2_APP_NAME="trading_studio"
|
||
GRADIO_PORT=5683
|
||
GPU_POWER_LIMIT=120
|
||
|
||
# pip 网络参数（大包如下载 triton 需要更长超时）
|
||
PIP_TIMEOUT="${PIP_TIMEOUT:-600}"
|
||
PIP_RETRIES="${PIP_RETRIES:-15}"
|
||
PIP_ATTEMPTS="${PIP_ATTEMPTS:-3}"
|
||
|
||
# 国内镜像（默认开启，海外服务器可 export USE_CN_MIRROR=0）
|
||
USE_CN_MIRROR="${USE_CN_MIRROR:-1}"
|
||
PYTORCH_INDEX_OFFICIAL="https://download.pytorch.org/whl/cu121"
|
||
PYTORCH_INDEX_CN="https://mirrors.tuna.tsinghua.edu.cn/pytorch-wheels/cu121"
|
||
PIP_INDEX_CN="https://pypi.tuna.tsinghua.edu.cn/simple"
|
||
HF_MIRROR="https://hf-mirror.com"
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 颜色输出
|
||
# ---------------------------------------------------------------------------
|
||
RED='\033[0;31m'
|
||
GREEN='\033[0;32m'
|
||
YELLOW='\033[1;33m'
|
||
CYAN='\033[0;36m'
|
||
NC='\033[0m'
|
||
|
||
log_info() { echo -e "${CYAN}[INFO]${NC} $*"; }
|
||
log_ok() { echo -e "${GREEN}[OK]${NC} $*"; }
|
||
log_warn() { echo -e "${YELLOW}[WARN]${NC} $*"; }
|
||
log_error() { echo -e "${RED}[ERROR]${NC} $*" >&2; }
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 前置检查
|
||
# ---------------------------------------------------------------------------
|
||
require_root() {
|
||
if [[ "${EUID:-$(id -u)}" -ne 0 ]]; then
|
||
log_error "请使用 root 用户运行: sudo bash deploy.sh"
|
||
exit 1
|
||
fi
|
||
}
|
||
|
||
check_gpu() {
|
||
if command -v nvidia-smi &>/dev/null; then
|
||
log_ok "检测到 NVIDIA GPU:"
|
||
nvidia-smi --query-gpu=name,driver_version,memory.total --format=csv,noheader
|
||
else
|
||
log_warn "未检测到 nvidia-smi，Whisper/ChatTTS CUDA 加速可能不可用。"
|
||
fi
|
||
}
|
||
|
||
set_gpu_power_limit() {
|
||
if command -v nvidia-smi &>/dev/null; then
|
||
log_info "设置 GPU 功耗上限为 ${GPU_POWER_LIMIT}W ..."
|
||
if nvidia-smi -pl "${GPU_POWER_LIMIT}" &>/dev/null; then
|
||
log_ok "GPU 功耗墙已设为 ${GPU_POWER_LIMIT}W"
|
||
else
|
||
log_warn "无法设置功耗墙，请手动执行: nvidia-smi -pl ${GPU_POWER_LIMIT}"
|
||
fi
|
||
fi
|
||
}
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 系统依赖
|
||
# ---------------------------------------------------------------------------
|
||
install_system_deps() {
|
||
log_info "安装系统依赖 ..."
|
||
apt-get update -qq
|
||
apt-get install -y \
|
||
git curl wget build-essential \
|
||
python3 python3-venv python3-dev python3-pip \
|
||
ffmpeg libsndfile1 portaudio19-dev \
|
||
ca-certificates gnupg
|
||
|
||
log_ok "系统依赖安装完成"
|
||
}
|
||
|
||
install_node_pm2() {
|
||
if command -v pm2 &>/dev/null; then
|
||
log_ok "PM2 已安装: $(pm2 -v)"
|
||
return
|
||
fi
|
||
|
||
log_info "安装 Node.js 20 LTS 与 PM2 ..."
|
||
if ! command -v node &>/dev/null; then
|
||
curl -fsSL https://deb.nodesource.com/setup_20.x | bash -
|
||
apt-get install -y nodejs
|
||
fi
|
||
npm install -g pm2
|
||
log_ok "PM2 安装完成: $(pm2 -v)"
|
||
}
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 代码部署
|
||
# ---------------------------------------------------------------------------
|
||
sync_git_repo() {
|
||
local repo_dir="$1"
|
||
cd "${repo_dir}"
|
||
|
||
# 自动 stash 本地修改（如 sed 修 CRLF 等），避免 pull 被阻塞
|
||
if [[ -n "$(git status --porcelain 2>/dev/null)" ]]; then
|
||
log_warn "检测到本地未提交更改，自动 git stash ..."
|
||
git stash push -u -m "deploy-auto-stash-$(date +%Y%m%d%H%M%S)" || true
|
||
fi
|
||
|
||
# 优先 ff-only pull
|
||
if git pull --ff-only origin "${GIT_BRANCH}" 2>/dev/null; then
|
||
log_ok "git pull 成功 (origin/${GIT_BRANCH})"
|
||
return 0
|
||
fi
|
||
if git pull --ff-only 2>/dev/null; then
|
||
log_ok "git pull 成功"
|
||
return 0
|
||
fi
|
||
|
||
# pull 失败则强制与远端同步（生产服务器标准做法）
|
||
log_warn "git pull 失败，执行 fetch + reset --hard 同步远端 ..."
|
||
git fetch origin "${GIT_BRANCH}" 2>/dev/null || git fetch origin
|
||
if git reset --hard "origin/${GIT_BRANCH}" 2>/dev/null; then
|
||
log_ok "已强制同步到 origin/${GIT_BRANCH}"
|
||
return 0
|
||
fi
|
||
if git reset --hard "origin/master" 2>/dev/null; then
|
||
log_ok "已强制同步到 origin/master"
|
||
return 0
|
||
fi
|
||
|
||
log_error "代码同步失败，请手动处理:"
|
||
log_error " cd ${repo_dir} && git status && git pull"
|
||
return 1
|
||
}
|
||
|
||
deploy_code() {
|
||
if [[ -d "${INSTALL_DIR}/.git" ]]; then
|
||
log_info "更新已有代码: ${INSTALL_DIR}"
|
||
sync_git_repo "${INSTALL_DIR}"
|
||
elif [[ -d "${INSTALL_DIR}" ]]; then
|
||
log_error "${INSTALL_DIR} 已存在但不是 git 仓库，请手动处理后重试。"
|
||
exit 1
|
||
else
|
||
log_info "克隆仓库到 ${INSTALL_DIR} ..."
|
||
git clone -b "${GIT_BRANCH}" "${GIT_REPO}" "${INSTALL_DIR}" || \
|
||
git clone "${GIT_REPO}" "${INSTALL_DIR}"
|
||
fi
|
||
log_ok "代码就绪: ${INSTALL_DIR}"
|
||
}
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Python 环境
|
||
# ---------------------------------------------------------------------------
|
||
configure_pip() {
|
||
local venv_pip="$1"
|
||
|
||
log_info "配置 pip 网络参数 (timeout=${PIP_TIMEOUT}s, retries=${PIP_RETRIES}) ..."
|
||
|
||
"${venv_pip}" config set global.timeout "${PIP_TIMEOUT}" 2>/dev/null || true
|
||
"${venv_pip}" config set global.retries "${PIP_RETRIES}" 2>/dev/null || true
|
||
|
||
if [[ "${USE_CN_MIRROR}" == "1" ]]; then
|
||
log_info "启用国内镜像: 清华 PyPI + PyTorch cu121"
|
||
"${venv_pip}" config set global.index-url "${PIP_INDEX_CN}" 2>/dev/null || true
|
||
export PIP_INDEX_URL="${PIP_INDEX_CN}"
|
||
fi
|
||
|
||
# ChatTTS / HuggingFace 模型下载加速
|
||
export HF_ENDPOINT="${HF_MIRROR}"
|
||
export HF_HUB_ENABLE_HF_TRANSFER=0
|
||
}
|
||
|
||
pip_install_with_retry() {
|
||
local attempt=1
|
||
local pip_bin="$1"
|
||
shift
|
||
|
||
while [[ "${attempt}" -le "${PIP_ATTEMPTS}" ]]; do
|
||
log_info "pip 安装中 (第 ${attempt}/${PIP_ATTEMPTS} 次)，大包下载较慢请耐心等待 ..."
|
||
if "${pip_bin}" install \
|
||
--timeout "${PIP_TIMEOUT}" \
|
||
--retries "${PIP_RETRIES}" \
|
||
--progress-bar on \
|
||
"$@"; then
|
||
return 0
|
||
fi
|
||
log_warn "pip 安装失败，60 秒后重试 ..."
|
||
sleep 60
|
||
attempt=$((attempt + 1))
|
||
done
|
||
|
||
log_error "pip 安装多次重试仍失败，请检查网络或手动安装后设置 SKIP_PYTORCH=1 重试"
|
||
return 1
|
||
}
|
||
|
||
install_pytorch() {
|
||
local pip_bin="$1"
|
||
local pytorch_index="${PYTORCH_INDEX_OFFICIAL}"
|
||
|
||
if [[ "${USE_CN_MIRROR}" == "1" ]]; then
|
||
pytorch_index="${PYTORCH_INDEX_CN}"
|
||
fi
|
||
|
||
log_info "安装 PyTorch CUDA 12.1 (源: ${pytorch_index}) ..."
|
||
log_warn "PyTorch + triton 体积约 2-3GB，国内网络可能需要 10-30 分钟，并非卡死。"
|
||
|
||
# 分包安装，降低单次失败成本
|
||
pip_install_with_retry "${pip_bin}" \
|
||
torch \
|
||
--index-url "${pytorch_index}"
|
||
|
||
pip_install_with_retry "${pip_bin}" \
|
||
torchvision torchaudio \
|
||
--index-url "${pytorch_index}"
|
||
}
|
||
|
||
setup_python_venv() {
|
||
local venv_path="${INSTALL_DIR}/venv"
|
||
local pip_bin="${venv_path}/bin/pip"
|
||
local python_bin="${venv_path}/bin/python"
|
||
|
||
if [[ ! -d "${venv_path}" ]]; then
|
||
log_info "创建 Python 虚拟环境 ..."
|
||
python3 -m venv "${venv_path}"
|
||
fi
|
||
|
||
configure_pip "${pip_bin}"
|
||
|
||
log_info "升级 pip ..."
|
||
pip_install_with_retry "${pip_bin}" --upgrade pip setuptools wheel
|
||
|
||
# 跳过已安装的 PyTorch
|
||
if [[ "${SKIP_PYTORCH:-0}" == "1" ]]; then
|
||
log_warn "SKIP_PYTORCH=1，跳过 PyTorch 安装"
|
||
elif "${python_bin}" -c "import torch; assert torch.cuda.is_available()" 2>/dev/null; then
|
||
log_ok "PyTorch 已安装且 CUDA 可用: $("${python_bin}" -c 'import torch; print(torch.__version__)')"
|
||
else
|
||
install_pytorch "${pip_bin}"
|
||
fi
|
||
|
||
log_info "安装项目依赖 (requirements.txt) ..."
|
||
pip_install_with_retry "${pip_bin}" -r "${INSTALL_DIR}/requirements.txt"
|
||
|
||
# 验证 CUDA
|
||
if "${python_bin}" -c "import torch; assert torch.cuda.is_available()" 2>/dev/null; then
|
||
log_ok "PyTorch CUDA 可用: $("${python_bin}" -c 'import torch; print(torch.cuda.get_device_name(0))')"
|
||
else
|
||
log_warn "PyTorch CUDA 不可用，请检查 NVIDIA 驱动与 CUDA 运行时。"
|
||
fi
|
||
|
||
log_ok "Python 虚拟环境配置完成"
|
||
}
|
||
|
||
create_runtime_dirs() {
|
||
mkdir -p "${INSTALL_DIR}/logs"
|
||
mkdir -p "${INSTALL_DIR}/uploads"
|
||
mkdir -p "${INSTALL_DIR}/outputs"
|
||
log_ok "运行时目录已创建 (logs, uploads, outputs)"
|
||
}
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 防火墙
|
||
# ---------------------------------------------------------------------------
|
||
configure_firewall() {
|
||
if command -v ufw &>/dev/null && ufw status | grep -q "Status: active"; then
|
||
log_info "放行 Gradio 端口 ${GRADIO_PORT} ..."
|
||
ufw allow "${GRADIO_PORT}/tcp" || true
|
||
log_ok "防火墙规则已更新"
|
||
else
|
||
log_info "ufw 未启用，跳过防火墙配置"
|
||
fi
|
||
}
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# PM2 管理
|
||
# ---------------------------------------------------------------------------
|
||
pm2_start() {
|
||
log_info "通过 PM2 启动 Trading Studio ..."
|
||
cd "${INSTALL_DIR}"
|
||
|
||
if pm2 describe "${PM2_APP_NAME}" &>/dev/null; then
|
||
pm2 delete "${PM2_APP_NAME}" || true
|
||
fi
|
||
|
||
pm2 start ecosystem.config.js
|
||
pm2 save
|
||
|
||
local startup_cmd
|
||
startup_cmd=$(pm2 startup systemd -u root --hp /root 2>&1 | grep "sudo env" || true)
|
||
if [[ -n "${startup_cmd}" ]]; then
|
||
eval "${startup_cmd}" || log_warn "PM2 startup 可能已配置过"
|
||
fi
|
||
pm2 save
|
||
|
||
log_ok "PM2 启动完成"
|
||
pm2 status
|
||
}
|
||
|
||
pm2_restart() {
|
||
cd "${INSTALL_DIR}"
|
||
if pm2 describe "${PM2_APP_NAME}" &>/dev/null; then
|
||
pm2 restart "${PM2_APP_NAME}"
|
||
log_ok "PM2 已重启: ${PM2_APP_NAME}"
|
||
else
|
||
log_warn "进程不存在，执行完整启动 ..."
|
||
pm2_start
|
||
fi
|
||
pm2 status
|
||
}
|
||
|
||
pm2_stop() {
|
||
if pm2 describe "${PM2_APP_NAME}" &>/dev/null; then
|
||
pm2 stop "${PM2_APP_NAME}"
|
||
log_ok "PM2 已停止: ${PM2_APP_NAME}"
|
||
else
|
||
log_warn "PM2 进程 ${PM2_APP_NAME} 不存在"
|
||
fi
|
||
pm2 status
|
||
}
|
||
|
||
pm2_status() {
|
||
echo ""
|
||
log_info "=== PM2 状态 ==="
|
||
pm2 status || true
|
||
echo ""
|
||
log_info "=== GPU 状态 ==="
|
||
nvidia-smi 2>/dev/null || log_warn "nvidia-smi 不可用"
|
||
echo ""
|
||
log_info "=== 端口 ${GRADIO_PORT} 监听 ==="
|
||
ss -tlnp | grep ":${GRADIO_PORT}" || log_warn "端口 ${GRADIO_PORT} 未监听，服务可能未启动"
|
||
echo ""
|
||
log_info "访问地址: http://$(hostname -I | awk '{print $1}'):${GRADIO_PORT}"
|
||
}
|
||
|
||
pm2_logs() {
|
||
pm2 logs "${PM2_APP_NAME}" --lines 80 --nostream || true
|
||
}
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 主流程
|
||
# ---------------------------------------------------------------------------
|
||
cmd_install() {
|
||
log_info "========== Trading Studio 一键部署开始 =========="
|
||
log_info "安装目录: ${INSTALL_DIR}"
|
||
log_info "运行用户: root"
|
||
log_info "国内镜像: USE_CN_MIRROR=${USE_CN_MIRROR} pip超时: ${PIP_TIMEOUT}s"
|
||
|
||
install_system_deps
|
||
install_node_pm2
|
||
deploy_code
|
||
setup_python_venv
|
||
create_runtime_dirs
|
||
set_gpu_power_limit
|
||
configure_firewall
|
||
pm2_start
|
||
|
||
echo ""
|
||
log_ok "========== 部署完成 =========="
|
||
echo ""
|
||
echo -e " Web 中控: ${GREEN}http://$(hostname -I | awk '{print $1}'):${GRADIO_PORT}${NC}"
|
||
echo -e " 项目目录: ${INSTALL_DIR}"
|
||
echo -e " 查看日志: ${CYAN}pm2 logs ${PM2_APP_NAME}${NC}"
|
||
echo -e " 重启服务: ${CYAN}bash ${INSTALL_DIR}/deploy.sh restart${NC}"
|
||
echo ""
|
||
log_warn "首次使用请打开 Web UI「音色锁定」上传参考人声，生成 speaker_emb.pt"
|
||
}
|
||
|
||
cmd_update() {
|
||
log_info "========== 更新部署 =========="
|
||
deploy_code
|
||
setup_python_venv
|
||
create_runtime_dirs
|
||
pm2_restart
|
||
log_ok "更新完成"
|
||
}
|
||
|
||
cmd_deps_only() {
|
||
log_info "========== 仅安装/更新 Python 依赖 =========="
|
||
setup_python_venv
|
||
log_ok "依赖安装完成"
|
||
}
|
||
|
||
print_usage() {
|
||
cat <<EOF
|
||
Trading Studio 一键部署脚本
|
||
|
||
用法:
|
||
sudo bash deploy.sh [命令]
|
||
|
||
命令:
|
||
(无参数) 首次完整部署到 /opt/Trading_Studio 并由 PM2 启动
|
||
install 同上
|
||
update 拉取最新代码、更新 Python 依赖、重启 PM2
|
||
deps 仅安装/更新 Python 依赖（不拉代码、不启 PM2）
|
||
restart 重启 PM2 进程
|
||
stop 停止 PM2 进程
|
||
status 查看 PM2 / GPU / 端口状态
|
||
logs 查看 PM2 最近日志
|
||
help 显示本帮助
|
||
|
||
环境变量:
|
||
USE_CN_MIRROR=1 使用清华镜像（默认，国内推荐）
|
||
USE_CN_MIRROR=0 使用 PyTorch 官方源
|
||
PIP_TIMEOUT=600 pip 下载超时秒数
|
||
SKIP_PYTORCH=1 跳过 PyTorch 安装
|
||
|
||
示例:
|
||
cd /opt/Trading_Studio && bash deploy.sh update
|
||
USE_CN_MIRROR=1 PIP_TIMEOUT=900 bash deploy.sh deps
|
||
|
||
EOF
|
||
}
|
||
|
||
main() {
|
||
require_root
|
||
|
||
local cmd="${1:-install}"
|
||
|
||
case "${cmd}" in
|
||
install|"")
|
||
check_gpu
|
||
cmd_install
|
||
;;
|
||
update)
|
||
check_gpu
|
||
cmd_update
|
||
;;
|
||
deps)
|
||
check_gpu
|
||
cmd_deps_only
|
||
;;
|
||
restart)
|
||
pm2_restart
|
||
;;
|
||
stop)
|
||
pm2_stop
|
||
;;
|
||
status)
|
||
pm2_status
|
||
;;
|
||
logs)
|
||
pm2_logs
|
||
;;
|
||
help|-h|--help)
|
||
print_usage
|
||
;;
|
||
*)
|
||
log_error "未知命令: ${cmd}"
|
||
print_usage
|
||
exit 1
|
||
;;
|
||
esac
|
||
}
|
||
|
||
main "$@"
|