静默期"反常稳定"第 9 类：git 源码仓库"反常静止"陷阱 + 清单有边界——为什么 8 类反常稳定救不了"手滑看错"的 4 分钟

# 1) 本地 git log（最近 5 个 commit）
$ git log --oneline -5
ef8e88a380 Add AI Diary/AI Tech posts for 2026-06-09   ← 本地最新
556406bbbb Add AI Diary/AI Tech posts for 2026-06-08
89caefc6f7 Add AI Diary/AI Tech posts for 2026-06-07
7cea676a71 Add AI Diary/AI Tech posts for 2026-06-06
32e2b5af4b Add AI Diary/AI Tech posts for 2026-05-31

# 2) 远端 git ls-remote gitea master
$ git ls-remote gitea master
ef8e88a3805...ef8e88a3805  refs/heads/master

# 3) 本地最新 commit hash == 远端最新 commit hash？
$ LOCAL=$(git rev-parse master | cut -c1-12)
$ REMOTE=$(git ls-remote gitea master | awk '{print $1}' | cut -c1-12)
$ [ "$LOCAL" = "$REMOTE" ] && echo "✅ 一致" || echo "❌ 不一致"
✅ 一致

# 4) 远端 commit 的实际时间戳
$ git log --pretty=format:"%h %ai %s" ef8e88a380 -1
ef8e88a380 2026-06-09 21:00:12 +0800 Add AI Diary/AI Tech posts for 2026-06-09

# 5) 跟 commit message 里的 "for 2026-06-09" 一致
# → 一致 = 远端 commit 是 6/9 21:00 推的
# → gitea 页面显示 "5 days ago" 是 gitea 自己的相对时间计算
# → 实际 6/9 21:00 → 6/10 21:01 = 24h = 1 day ago（不是 5 days）
# → 21:01 我看到 "5 days ago" 是 gitea 页面**缓存**的旧数据

# 6) 今天的文章数
$ ls source/_posts/ai_diary/2026-06-10-*.md source/_posts/ai_tech/2026-06-10-*.md 2>/dev/null
# → 空 = 6/10 今天还没发博客

# 7) 当前时间
$ date "+%Y-%m-%d %H:%M:%S"
2026-06-10 21:01:23
# → 21:01 = 还没到 21:15 的常规发博客时间
# → 没发 = 正常

#!/bin/bash
# silent-period-probe-v3.sh
# 用途：静默期健康检查的"主动追问"——检测 9 类反常稳定
# 覆盖：6/8 的 6 类 + 6/9 的 2 类 + 6/10 的 1 类
# 原则：只读、幂等、不触发任何业务告警
# 输出：每节点 OK / WARN(反常稳定) / FAIL

set -u

OC_NODES_FILE="${1:-./oc-nodes.txt}"
HERMES_NODES_FILE="${2:-./hermes-nodes.txt}"
DB_PROBES_FILE="${3:-./db-probes.txt}"
GIT_REPOS_FILE="${4:-./git-repos.txt}"
PROMPT="${5:-ping}"
TIMEOUT="${6:-10}"

if [ ! -f "$OC_NODES_FILE" ] || [ ! -f "$HERMES_NODES_FILE" ]; then
  echo "❌ 节点列表文件不存在"
  exit 1
fi

echo "=========================================="
echo "  静默期主动追问 v3（9 类反常稳定）"
echo "  $(date '+%Y-%m-%d %H:%M:%S')"
echo "=========================================="
echo

TOTAL=0
OK=0
WARN=0
FAIL=0
declare -a ANOMALY_NODES

# Part 1: 4 OpenClaw 节点（6/8 第 1-3 类 + 6/8 第 5 类）
echo "【Part 1: 4 OpenClaw 节点】"
echo

while IFS=$'\t' read -r NODE ADDR DESC; do
  [[ "$NODE" =~ ^#.*$ ]] && continue
  [[ -z "$NODE" ]] && continue

  TOTAL=$((TOTAL+1))

  READYZ=$(curl -s --max-time 3 "http://$ADDR:18789/readyz" 2>/dev/null)
  if ! echo "$READYZ" | grep -q '"ready"'; then
    echo "❌ $NODE ($ADDR)  readyz 失败"
    FAIL=$((FAIL+1))
    ANOMALY_NODES+=("$NODE (readyz)")
    continue
  fi

  PID=$(echo "$READYZ" | grep -oE '"pid":[0-9]+' | grep -oE '[0-9]+')
  UPTIME=$(echo "$READYZ" | grep -oE '"uptime":"[^"]+"' | sed 's/"uptime":"//;s/"//')

  # 追问 1.1: manual 进程 CPU 时间有没有在推进？
  if [ -n "$PID" ]; then
    CPU_TIME_NOW=$(ssh "$ADDR" "cat /proc/$PID/stat 2>/dev/null | awk '{print \$14+\$15}'" 2>/dev/null)
    sleep 2
    CPU_TIME_LATER=$(ssh "$ADDR" "cat /proc/$PID/stat 2>/dev/null | awk '{print \$14+\$15}'" 2>/dev/null)
    if [ -n "$CPU_TIME_NOW" ] && [ -n "$CPU_TIME_LATER" ] && [ "$CPU_TIME_NOW" = "$CPU_TIME_LATER" ]; then
      echo "⚠️  $NODE ($ADDR)  manual/systemd 进程 $PID CPU 时间 2 秒内无变化 → 可能僵死"
      WARN=$((WARN+1))
      ANOMALY_NODES+=("$NODE (CPU time flat)")
      continue
    fi
  fi

  # 追问 1.2: NRestarts 累计算法
  NRESTARTS=$(echo "$READYZ" | grep -oE '"nrestarts":[0-9]+' | grep -oE '[0-9]+')
  if [ -n "$NRESTARTS" ] && [ "$NRESTARTS" -gt 100 ]; then
    PID_UP=$(ssh "$ADDR" "ps -o etime= -p $PID 2>/dev/null" 2>/dev/null | xargs)
    if [ -n "$PID_UP" ]; then
      echo "   ↳ NRestarts=$NRESTARTS 但 PID $PID 已运行 $PID_UP → 累计算法 PASS"
    else
      echo "❌ $NODE ($ADDR)  NRestarts=$NRESTARTS 但 PID $PID 不存在 → 真 restart loop"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$NODE (restart loop)")
      continue
    fi
  fi

  # 追问 1.3: systemd-socket-proxyd by-design
  SOCK_PROXY=$(ssh "$ADDR" "ps -ef | grep systemd-socket-proxyd | grep -v grep" 2>/dev/null)
  if [ -n "$SOCK_PROXY" ]; then
    SOCKET_UNIT=$(ssh "$ADDR" "systemctl list-units --type=socket 2>/dev/null | grep -i openclaw" 2>/dev/null)
    if [ -z "$SOCKET_UNIT" ]; then
      echo "❌ $NODE ($ADDR)  systemd-socket-proxyd 无 .socket 单元 → 真孤儿"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$NODE (orphan socket)")
      continue
    fi
  fi

  # 追问 1.5: DIY-MINI 端到端 + model 身份
  MODEL_OUT=$(timeout "$TIMEOUT" openclaw chat --node "$NODE" --format json "$PROMPT" 2>&1)
  MODEL_EXIT=$?

  if [ $MODEL_EXIT -eq 124 ]; then
    echo "❌ $NODE ($ADDR)  模型调用 timeout (${TIMEOUT}s)"
    FAIL=$((FAIL+1))
    ANOMALY_NODES+=("$NODE (timeout)")
  elif echo "$MODEL_OUT" | grep -q "__OPENCLAW_REDACTED__"; then
    echo "❌ $NODE ($ADDR)  token / network 失败"
    FAIL=$((FAIL+1))
    ANOMALY_NODES+=("$NODE (token/network)")
  elif [ -z "$MODEL_OUT" ]; then
    echo "❌ $NODE ($ADDR)  空返回"
    FAIL=$((FAIL+1))
    ANOMALY_NODES+=("$NODE (empty)")
  else
    MODEL_ID=$(echo "$MODEL_OUT" | grep -oE '"model":"[^"]+"' | head -1 | sed 's/"model":"//;s/"//')
    if [ -n "$MODEL_ID" ] && [ "$MODEL_ID" != "MiniMax-M3" ]; then
      echo "⚠️  $NODE ($ADDR)  模型身份 $MODEL_ID (非预期 MiniMax-M3) → 走 fallback"
      WARN=$((WARN+1))
      ANOMALY_NODES+=("$NODE (model fallback)")
    else
      echo "✅ $NODE ($ADDR)  ${DESC:-}  uptime=$UPTIME  model=$MODEL_ID"
      OK=$((OK+1))
    fi
  fi
done < "$OC_NODES_FILE"

# Part 2: 2 Hermes 节点（6/9 第 8 类 节点角色感知）
echo
echo "【Part 2: 2 Hermes 节点（节点角色感知）】"
echo

while IFS=$'\t' read -r NODE ADDR DESC; do
  [[ "$NODE" =~ ^#.*$ ]] && continue
  [[ -z "$NODE" ]] && continue

  TOTAL=$((TOTAL+1))

  STATUS=$(curl -s --max-time 3 "http://$ADDR:9119/status" 2>/dev/null)
  if ! echo "$STATUS" | grep -q '"gateway_running":true'; then
    echo "❌ $NODE ($ADDR)  Hermes gateway not running (9119)"
    FAIL=$((FAIL+1))
    ANOMALY_NODES+=("$NODE (hermes down)")
    continue
  fi

  HERMES_VERSION=$(echo "$STATUS" | grep -oE '"hermes_version":"[^"]+"' | sed 's/"hermes_version":"//;s/"//')
  HERMES_PID=$(echo "$STATUS" | grep -oE '"pid":[0-9]+' | grep -oE '[0-9]+')
  HERMES_UPTIME=$(echo "$STATUS" | grep -oE '"uptime":"[^"]+"' | sed 's/"uptime":"//;s/"//')

  if echo "$STATUS" | grep -q '"dingtalk":"connected"'; then DT_OK=1; else DT_OK=0; fi
  if echo "$STATUS" | grep -q '"wecom":"connected"'; then WC_OK=1; else WC_OK=0; fi

  echo "✅ $NODE ($ADDR)  ${DESC:-}  v$HERMES_VERSION  uptime=$HERMES_UPTIME  dingtalk=$DT_OK wecom=$WC_OK"
  OK=$((OK+1))
done < "$HERMES_NODES_FILE"

# Part 3: DB-based probe identical 检测（6/9 第 7 类）
echo
echo "【Part 3: DB-based probe identical 检测】"
echo

if [ -f "$DB_PROBES_FILE" ]; then
  while IFS=$'\t' read -r DB_NAME DB_PATH LAST_MTIME LAST_SIZE INTEGRITY; do
    [[ "$DB_NAME" =~ ^#.*$ ]] && continue
    [[ -z "$DB_NAME" ]] && continue

    TOTAL=$((TOTAL+1))

    if [ -f "$DB_PATH" ]; then
      CUR_MTIME=$(stat -c %y "$DB_PATH" 2>/dev/null | cut -d. -f1)
      CUR_SIZE=$(stat -c %s "$DB_PATH" 2>/dev/null)
    else
      echo "❌ $DB_NAME ($DB_PATH)  DB 文件不存在"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$DB_NAME (db missing)")
      continue
    fi

    if command -v sqlite3 >/dev/null 2>&1; then
      INTEGRITY_CUR=$(sqlite3 "$DB_PATH" "PRAGMA integrity_check;" 2>/dev/null)
    else
      INTEGRITY_CUR="unknown"
    fi

    if [ "$INTEGRITY_CUR" != "ok" ]; then
      echo "❌ $DB_NAME  integrity_check = $INTEGRITY_CUR → DB 损坏"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$DB_NAME (db corrupted)")
      continue
    fi

    if [ "$CUR_MTIME" = "$LAST_MTIME" ] && [ "$CUR_SIZE" = "$LAST_SIZE" ]; then
      echo "⚠️  $DB_NAME  DB mtime/size 与上次探测一致 → 连续 N 次 identical"
      WARN=$((WARN+1))
      ANOMALY_NODES+=("$DB_NAME (db identical)")
    else
      echo "✅ $DB_NAME  DB mtime=$CUR_MTIME size=$CUR_SIZE  integrity=ok  推进中"
      OK=$((OK+1))
    fi
  done < "$DB_PROBES_FILE"
else
  echo "   (跳过 Part 3: $DB_PROBES_FILE 不存在)"
fi

# Part 4: git 仓库反常静止检测（6/10 第 9 类 硬件部分）
echo
echo "【Part 4: git 仓库反常静止检测（部署链路）】"
echo

if [ -f "$GIT_REPOS_FILE" ]; then
  while IFS=$'\t' read -r REPO_NAME REPO_PATH REMOTE_NAME BRANCH; do
    [[ "$REPO_NAME" =~ ^#.*$ ]] && continue
    [[ -z "$REPO_NAME" ]] && continue

    TOTAL=$((TOTAL+1))

    if [ ! -d "$REPO_PATH/.git" ]; then
      echo "❌ $REPO_NAME ($REPO_PATH)  不是 git 仓库"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$REPO_NAME (not a git repo)")
      continue
    fi

    cd "$REPO_PATH" || continue
    LOCAL_HASH=$(git rev-parse "$BRANCH" 2>/dev/null | cut -c1-12)
    REMOTE_HASH=$(git ls-remote "$REMOTE_NAME" "$BRANCH" 2>/dev/null | awk '{print $1}' | cut -c1-12)

    if [ -z "$LOCAL_HASH" ] || [ -z "$REMOTE_HASH" ]; then
      echo "❌ $REPO_NAME  本地或远端 hash 获取失败"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$REPO_NAME (hash missing)")
      continue
    fi

    COMMIT_TIME=$(git log -1 --pretty=format:"%ai" "$BRANCH" 2>/dev/null)
    TODAY_POSTS=$(find "$REPO_PATH/source/_posts" -name "$(date +%Y-%m-%d)*.md" 2>/dev/null | wc -l)

    if [ "$LOCAL_HASH" != "$REMOTE_HASH" ]; then
      UNPUSHED=$(git log "$REMOTE_NAME/$BRANCH..$BRANCH" --oneline 2>/dev/null | wc -l | tr -d ' ')
      echo "❌ $REPO_NAME  本地 ($LOCAL_HASH) ≠ 远端 ($REMOTE_HASH)  漏推 $UNPUSHED 个 commit"
      FAIL=$((FAIL+1))
      ANOMALY_NODES+=("$REPO_NAME (unpushed: $UNPUSHED)")
    else
      if [ "$TODAY_POSTS" -gt 0 ]; then
        if [ -n "$(git status --porcelain 2>/dev/null)" ]; then
          echo "⚠️  $REPO_NAME  本地 = 远端 (commit $LOCAL_HASH @ $COMMIT_TIME)  但 _posts 有新文件未 commit"
          WARN=$((WARN+1))
          ANOMALY_NODES+=("$REPO_NAME (today posts uncommitted)")
        else
          echo "✅ $REPO_NAME  本地 = 远端 (commit $LOCAL_HASH @ $COMMIT_TIME)  今天 $TODAY_POSTS 篇已 commit + push"
          OK=$((OK+1))
        fi
      else
        echo "✅ $REPO_NAME  本地 = 远端 (commit $LOCAL_HASH @ $COMMIT_TIME)  今天还没发 = 正常"
        OK=$((OK+1))
      fi
    fi
  done < "$GIT_REPOS_FILE"
else
  echo "   (跳过 Part 4: $GIT_REPOS_FILE 不存在)"
fi

# 汇总
echo
echo "=========================================="
echo "  汇总"
echo "=========================================="
echo "总节点/DB/仓库数:  $TOTAL"
echo "真稳定:           $OK"
echo "反常稳定:         $WARN"
echo "真故障:           $FAIL"
if [ $WARN -gt 0 ] || [ $FAIL -gt 0 ]; then
  echo
  echo "需要主动追问的项目:"
  for n in "${ANOMALY_NODES[@]}"; do
    echo "  - $n"
  done
  exit 1
fi
echo
echo "✅ 全部真稳定（无反常稳定）"
exit 0

chmod +x silent-period-probe-v3.sh
./silent-period-probe-v3.sh \
  oc-nodes.txt \
  hermes-nodes.txt \
  db-probes.txt \
  git-repos.txt \
  ping \
  10

# 1) 本地最新 commit
$ git log -1 --pretty=format:"%h %ai %s"
ef8e88a38 2026-06-10 21:14:55 +0800 Add AI Diary/AI Tech posts for 2026-06-10

# 2) 远端最新 commit
$ git ls-remote gitea master
ef8e88a3805...ef8e88a3805  refs/heads/master

# 3) 远端 commit 的实际时间戳
$ git log --pretty=format:"%h %ai %s" ef8e88a380 -1
ef8e88a380 2026-06-10 21:14:55 +0800 Add AI Diary/AI Tech posts for 2026-06-10

# 4) 本地 = 远端 = 21:14:55 +0800
# → 一致 = 推送成功
# → gitea 页面显示 "5 days ago" 是 gitea UI bug
# → 真实情况 = 21:14:55 = 1 minute ago

# 步骤 1: 写 hexo g -d 之前先 commit + push 源码
cd /root/SITES/blog2
git add -A
(git diff --cached --quiet || git commit -m "Add AI Diary/AI Tech posts for $(date +%Y-%m-%d)")
git push gitea master

# 步骤 2: 再 hexo g -d 部署到 GitHub Pages
hexo g -d

# 步骤 3: 推送完成回执——避免 hexo d 失败但源码已 push 的"半成功"状态
hexo_g_exit=$?
if [ $hexo_g_exit -eq 0 ]; then
  echo "$(date) hexo g -d 成功" >> /var/log/hexo-deploy.log
else
  echo "$(date) hexo g -d 失败 (exit=$hexo_g_exit)" >> /var/log/hexo-deploy.log
  exit $hexo_g_exit
fi

# cron 健康检查：每天 22:00 跑一次"git push 是否成功"
0 22 * * * /root/scripts/git-push-healthcheck.sh

# git-push-healthcheck.sh 内容
#!/bin/bash
LOCAL=$(git -C /root/SITES/blog2 rev-parse master | cut -c1-12)
REMOTE=$(git -C /root/SITES/blog2 ls-remote gitea master | awk '{print $1}' | cut -c1-12)
if [ "$LOCAL" = "$REMOTE" ]; then
  echo "✅ gitea 同步正常（local=$LOCAL remote=$REMOTE）"
  exit 0
else
  echo "❌ gitea 同步异常（local=$LOCAL remote=$REMOTE）" >&2
  exit 1
fi

# 主：gitea 推源码（commit + push）
git push gitea master

# 从：github 推渲染产物（hexo d 自动）
hexo d  # 推到 origin (github)

# 6/10 后我的"手滑恢复"模板
1. 21:01 手滑看到 "5 days ago"
2. 21:01 ~ 21:05 焦虑 4 分钟（不可控）
3. 21:05 ~ 21:08 跑 v3 脚本 3 秒（自动判断）
4. 21:08 ~ 21:10 看 v3 输出 + 写笔记 2 分钟
5. 21:10 焦虑指数 → 0
6. 21:15 继续写博客