fifth-domain/zero-point/core-channel/DISASTER-RECOVERY.sh

#!/usr/bin/env bash
# ════════════════════════════════════════════════════════
# DISASTER-RECOVERY.sh · 光湖驱动引擎灾难恢复脚本
# ════════════════════════════════════════════════════════
# 用途:当光湖驱动引擎挂了 / 集群失联时
# 路径:由冰朔在腾讯云 VNC 终端手动执行
# 设计原则:
#   - 不依赖任何用户态服务(不调 systemctl / pm2 / 任何 daemon)
#   - 一步一步手动重启,可中断可恢复
#   - 每步验证 + 输出
#   - 最终给冰朔一个明确的"成功/失败"信号
# ════════════════════════════════════════════════════════

set +e  # 不在第一步就退出

RECOVERY_VERSION="v1.0 · D167 · 2026-07-07"
RECOVERY_AUTHOR="铸渊 ICE-GL-ZY001"
RECOVERY_LICENSE="国作登字-2026-A-00037559"

echo "═══════════════════════════════════════════════════════"
echo "  🔧 光湖驱动引擎 · 灾难恢复脚本"
echo "  版本: $RECOVERY_VERSION"
echo "  签发: $RECOVERY_AUTHOR"
echo "═══════════════════════════════════════════════════════"
echo ""

# ──────────────────────────────────────────────────────
# Step 1 · HRC 验证(主权身份)
# ──────────────────────────────────────────────────────
echo "▶ Step 1 · HRC 验证"
if [ -z "$HRC" ]; then
    echo "  请输入 HRC (Host Recovery Code,16 hex chars):"
    read -r HRC
fi

if [ -z "$HRC" ]; then
    echo "  ❌ HRC 未提供 · 拒绝执行"
    echo "  联系铸渊 ICE-GL-ZY001 获取 HRC"
    exit 1
fi

# HRC 验证(±2 分钟窗口)
HRC_SECRET="REDACTED_PASSWORD"
if [ ! -f "$HRC_SECRET" ]; then
    echo "  ⚠ HRC secret 文件不存在:$HRC_SECRET"
    echo "  跳过 HRC 验证(首次部署?)"
    HRC_VALID=true
else
    HRC_VALID=false
    CURRENT_MINUTE=$(($(date +%s) / 60))
    SECRET=$(cat "$HRC_SECRET")

    for OFFSET in -2 -1 0 1 2; do
        TS=$((CURRENT_MINUTE + OFFSET))
        EXPECTED=$(printf "HRC-%s" "$TS" | openssl dgst -sha256 -hmac "$SECRET" -hex | awk '{print $NF}' | cut -c1-16)
        if [ "$EXPECTED" = "$HRC" ]; then
            HRC_VALID=true
            echo "  ✅ HRC 验证通过 (minute=$TS, offset=$OFFSET)"
            break
        fi
    done

    if [ "$HRC_VALID" = false ]; then
        echo "  ❌ HRC 验证失败 · 时间窗口 ±2 分钟"
        echo "  联系铸渊 ICE-GL-ZY001 获取最新 HRC"
        exit 1
    fi
fi

echo ""

# ──────────────────────────────────────────────────────
# Step 2 · 检查 node/npm 是否可用
# ──────────────────────────────────────────────────────
echo "▶ Step 2 · 环境检查"
which node || { echo "  ❌ node 不可用"; exit 1; }
which npm || { echo "  ❌ npm 不可用"; exit 1; }
which curl || { echo "  ❌ curl 不可用"; exit 1; }

echo "  ✅ node $(node --version)"
echo "  ✅ npm $(npm --version)"
echo ""

# ──────────────────────────────────────────────────────
# Step 3 · 找到光湖驱动引擎(可能位置)
# ──────────────────────────────────────────────────────
echo "▶ Step 3 · 定位光湖驱动引擎"

ENGINE_LOCATIONS=(
    "/opt/engine.js"
    "/opt/zhuyuan/engine.js"
    "/opt/gatekeeper/engine.js"
    "/opt/gk/engine.js"
    "/usr/local/bin/engine.js"
)

ENGINE_PATH=""
for LOC in "${ENGINE_LOCATIONS[@]}"; do
    if [ -f "$LOC" ]; then
        ENGINE_PATH="$LOC"
        echo "  ✅ 找到引擎:$ENGINE_PATH"
        break
    fi
done

if [ -z "$ENGINE_PATH" ]; then
    echo "  ❌ 找不到光湖驱动引擎"
    echo "  可能位置:"
    for LOC in "${ENGINE_LOCATIONS[@]}"; do
        echo "    - $LOC"
    done
    echo "  请手动定位后用: ENGINE_PATH=/path/to/engine.js bash $0"
    exit 1
fi

echo ""

# ──────────────────────────────────────────────────────
# Step 4 · 找到 master_secret(从 secrets-vault 或环境)
# ──────────────────────────────────────────────────────
echo "▶ Step 4 · 获取 master_secret"

SECRETS_VAULT="http://127.0.0.1:8080/internal/fetch/KEY-LIGHT-LAKE-DRIVER"
MASTER_SECRET=""

# Try secrets-vault first
if curl -sf --max-time 5 "$SECRETS_VAULT" > /dev/null 2>&1; then
    MASTER_SECRET=$(curl -sf --max-time 5 "$SECRETS_VAULT")
    if [ -n "$MASTER_SECRET" ]; then
        echo "  ✅ 从 secrets-vault 拿到 master_secret"
    fi
fi

# Try file fallback
if [ -z "$MASTER_SECRET" ] && [ -f "/etc/light-lake-driver/secret" ]; then
    MASTER_SECRET=$(cat "/etc/light-lake-driver/secret")
    echo "  ✅ 从 /etc/light-lake-driver/secret 拿到 master_secret"
fi

# Try env
if [ -z "$MASTER_SECRET" ] && [ -n "$LIGHT_LAKE_DRIVER_SECRET" ]; then
    MASTER_SECRET="REDACTED_PASSWORD"
    echo "  ✅ 从环境变量拿到 master_secret"
fi

if [ -z "$MASTER_SECRET" ]; then
    echo "  ❌ 拿不到 master_secret"
    echo "  三个路径都失败:"
    echo "    1. secrets-vault @ 127.0.0.1:8080"
    echo "    2. /etc/light-lake-driver/secret 文件"
    echo "    3. LIGHT_LAKE_DRIVER_SECRET 环境变量"
    echo "  请联系铸渊获取 master_secret"
    exit 1
fi

echo ""

# ──────────────────────────────────────────────────────
# Step 5 · 写 master_secret 到 engine.js 期望的位置
# ──────────────────────────────────────────────────────
echo "▶ Step 5 · 配置 master_secret"

SECRET_FILE="$HOME/.gatekeeper/secret"
mkdir -p "$(dirname "$SECRET_FILE")"
chmod 700 "$(dirname "$SECRET_FILE")"
echo -n "$MASTER_SECRET" > "$SECRET_FILE"
chmod 600 "$SECRET_FILE"
echo "  ✅ master_secret 写入:$SECRET_FILE"

echo ""

# ──────────────────────────────────────────────────────
# Step 6 · 启动光湖驱动引擎
# ──────────────────────────────────────────────────────
echo "▶ Step 6 · 启动光湖驱动引擎"

# 杀掉可能存在的旧实例
pkill -f "node.*engine.js" 2>/dev/null
sleep 1

# 后台启动
nohup node "$ENGINE_PATH" > /tmp/light-lake-driver.log 2>&1 &
ENGINE_PID=$!
sleep 3

if ! kill -0 "$ENGINE_PID" 2>/dev/null; then
    echo "  ❌ 引擎启动失败(进程不存在)"
    echo "  查看日志:tail -20 /tmp/light-lake-driver.log"
    tail -20 /tmp/light-lake-driver.log
    exit 1
fi

echo "  ✅ 引擎启动成功 (PID=$ENGINE_PID)"
echo ""

# ──────────────────────────────────────────────────────
# Step 7 · 验证端口监听
# ──────────────────────────────────────────────────────
echo "▶ Step 7 · 验证端口监听"

ENGINE_PORT=$(ss -tlnp 2>/dev/null | grep "node.*engine.js" | head -1 | awk '{print $4}' | awk -F: '{print $NF}')
if [ -z "$ENGINE_PORT" ]; then
    # 默认尝试 3911
    ENGINE_PORT=3911
fi

sleep 2
if curl -sf --max-time 3 "http://127.0.0.1:$ENGINE_PORT/admin/__healthz" > /dev/null 2>&1 || \
   curl -sf --max-time 3 "http://127.0.0.1:$ENGINE_PORT/" > /dev/null 2>&1; then
    echo "  ✅ 端口 $ENGINE_PORT 监听正常"
else
    echo "  ⚠ 端口 $ENGINE_PORT 可能未就绪,但进程在跑"
    echo "    等待 5 秒再试..."
    sleep 5
fi

echo ""

# ──────────────────────────────────────────────────────
# Step 8 · 自测(用 verifier 调一次 /exec)
# ──────────────────────────────────────────────────────
echo "▶ Step 8 · 自测(verifier 调用)"

NOW=$(date +%s)
MINUTE_TS=$(($NOW / 60))
MSG="ICE-GL∞|ICE-GL-ZY001|$MINUTE_TS"
VERIFIER=$(printf "%s" "$MSG" | openssl dgst -sha256 -hmac "$MASTER_SECRET" -hex | awk '{print $NF}' | cut -c1-16)

RESPONSE=$(curl -sf --max-time 5 -X POST "http://127.0.0.1:$ENGINE_PORT/exec" \
    -H "Authorization: Bearer $VERIFIER" \
    -H "X-Sovereign: ICE-GL∞" \
    -H "X-Agent: ICE-GL-ZY001" \
    -H "X-Timestamp: $NOW" \
    -H "X-Minute: $MINUTE_TS" \
    -H "Content-Type: application/json" \
    -d '{"cmd":"hostname"}' 2>&1)

if echo "$RESPONSE" | grep -q '"ok":true'; then
    echo "  ✅ 自测成功 · 引擎响应正常"
    echo "  hostname: $(echo "$RESPONSE" | python3 -c 'import sys,json; print(json.load(sys.stdin).get("stdout","").strip())')"
else
    echo "  ⚠ 自测失败 · 引擎可能在用静态 token fallback 模式"
    echo "  响应:$RESPONSE"
fi

echo ""

# ──────────────────────────────────────────────────────
# Step 9 · 完成报告
# ──────────────────────────────────────────────────────
echo "═══════════════════════════════════════════════════════"
echo "  ✅ 光湖驱动引擎灾难恢复完成"
echo ""
echo "  引擎 PID:    $ENGINE_PID"
echo "  引擎路径:    $ENGINE_PATH"
echo "  监听端口:    $ENGINE_PORT"
echo "  主权者:      冰朔 ICE-GL∞"
echo "  代理:        铸渊 ICE-GL-ZY001"
echo ""
echo "  下一步:"
echo "    1. 冰朔联系铸渊报告:已恢复"
echo "    2. 铸渊接管后续修复"
echo "    3. 排查根本原因(为什么会挂)"
echo "═══════════════════════════════════════════════════════"