mgmt/scripts/utilities/consul-cluster-manager.sh

#!/bin/bash

# Consul 集群管理脚本
# 提供集群状态检查、重启、停止等功能

set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
INVENTORY_FILE="$PROJECT_ROOT/configuration/inventories/production/consul-cluster.ini"

# 颜色定义
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

# 打印带颜色的消息
print_status() {
    echo -e "${GREEN}[INFO]${NC} $1"
}

print_warning() {
    echo -e "${YELLOW}[WARN]${NC} $1"
}

print_error() {
    echo -e "${RED}[ERROR]${NC} $1"
}

print_header() {
    echo -e "${BLUE}=== $1 ===${NC}"
}

# 检查必要文件
check_prerequisites() {
    if [[ ! -f "$INVENTORY_FILE" ]]; then
        print_error "清单文件不存在: $INVENTORY_FILE"
        exit 1
    fi
    
    if ! command -v ansible &> /dev/null; then
        print_error "未找到 ansible 命令"
        exit 1
    fi
}

# 显示帮助信息
show_help() {
    echo "Consul 集群管理脚本"
    echo
    echo "用法: $0 [命令]"
    echo
    echo "命令:"
    echo "  status      - 检查集群状态"
    echo "  members     - 显示集群成员"
    echo "  leader      - 显示集群领导者"
    echo "  restart     - 重启 Consul 服务"
    echo "  stop        - 停止 Consul 服务"
    echo "  start       - 启动 Consul 服务"
    echo "  logs        - 查看服务日志"
    echo "  health      - 健康检查"
    echo "  cleanup     - 清理 Consul 数据（危险操作）"
    echo "  help        - 显示此帮助信息"
    echo
}

# 检查集群状态
check_status() {
    print_header "Consul 服务状态"
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "systemctl is-active consul" -o
    
    echo
    print_header "Consul 进程状态"
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "ps aux | grep consul | grep -v grep" -o
}

# 显示集群成员
show_members() {
    print_header "Consul 集群成员"
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "consul members" -o
}

# 显示集群领导者
show_leader() {
    print_header "Consul 集群领导者"
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "consul operator raft list-peers" -o
    
    echo
    print_header "通过 API 检查领导者"
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "curl -s http://localhost:8500/v1/status/leader" -o
}

# 重启服务
restart_service() {
    print_header "重启 Consul 服务"
    print_warning "即将重启所有 Consul 节点..."
    read -p "确认继续? (y/N): " confirm
    if [[ $confirm != "y" && $confirm != "Y" ]]; then
        print_status "操作已取消"
        return
    fi
    
    ansible -i "$INVENTORY_FILE" consul_cluster -m systemd -a "name=consul state=restarted" -b
    
    print_status "等待服务启动..."
    sleep 10
    check_status
}

# 停止服务
stop_service() {
    print_header "停止 Consul 服务"
    print_warning "即将停止所有 Consul 节点..."
    read -p "确认继续? (y/N): " confirm
    if [[ $confirm != "y" && $confirm != "Y" ]]; then
        print_status "操作已取消"
        return
    fi
    
    ansible -i "$INVENTORY_FILE" consul_cluster -m systemd -a "name=consul state=stopped" -b
}

# 启动服务
start_service() {
    print_header "启动 Consul 服务"
    ansible -i "$INVENTORY_FILE" consul_cluster -m systemd -a "name=consul state=started" -b
    
    print_status "等待服务启动..."
    sleep 10
    check_status
}

# 查看日志
show_logs() {
    print_header "Consul 服务日志"
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "journalctl -u consul --no-pager -n 20" -o
}

# 健康检查
health_check() {
    print_header "Consul 健康检查"
    
    # 检查服务状态
    print_status "检查服务状态..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "systemctl is-active consul" -o
    
    echo
    # 检查端口监听
    print_status "检查端口监听..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "ss -tlnp | grep :8500" -o
    
    echo
    # 检查集群成员
    print_status "检查集群成员..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "consul members | wc -l" -o
    
    echo
    # 检查 API 响应
    print_status "检查 API 响应..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "curl -s -o /dev/null -w '%{http_code}' http://localhost:8500/v1/status/leader" -o
}

# 清理数据（危险操作）
cleanup_data() {
    print_header "清理 Consul 数据"
    print_error "警告: 此操作将删除所有 Consul 数据，包括服务注册、KV 存储等！"
    print_error "此操作不可逆！"
    echo
    read -p "确认要清理所有数据? 请输入 'YES' 确认: " confirm
    if [[ $confirm != "YES" ]]; then
        print_status "操作已取消"
        return
    fi
    
    print_status "停止 Consul 服务..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m systemd -a "name=consul state=stopped" -b
    
    print_status "清理数据目录..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m shell -a "rm -rf /opt/consul/data/*" -b
    
    print_status "启动 Consul 服务..."
    ansible -i "$INVENTORY_FILE" consul_cluster -m systemd -a "name=consul state=started" -b
    
    print_status "数据清理完成"
}

# 主函数
main() {
    check_prerequisites
    
    case "${1:-help}" in
        status)
            check_status
            ;;
        members)
            show_members
            ;;
        leader)
            show_leader
            ;;
        restart)
            restart_service
            ;;
        stop)
            stop_service
            ;;
        start)
            start_service
            ;;
        logs)
            show_logs
            ;;
        health)
            health_check
            ;;
        cleanup)
            cleanup_data
            ;;
        help|--help|-h)
            show_help
            ;;
        *)
            print_error "未知命令: $1"
            echo
            show_help
            exit 1
            ;;
    esac
}

main "$@"