From 9ec0cacf45733bc8ff78601f75a7e7406e60a0c8 Mon Sep 17 00:00:00 2001
From: dekun <dekun@local>
Date: Wed, 29 Jul 2026 21:29:38 +0800
Subject: [PATCH] Improve apt-lock wait: stop unattended-upgrades during
 deploy.

Shows lock holders and optionally stops auto-upgrade so Node install is not stuck after reboot.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 deploy/lib/common.sh | 47 +++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 44 insertions(+), 3 deletions(-)

diff --git a/deploy/lib/common.sh b/deploy/lib/common.sh
index 81b5399..05b9b58 100755
--- a/deploy/lib/common.sh
+++ b/deploy/lib/common.sh
@@ -34,6 +34,37 @@ require_root() {
   fi
 }
 
+# 打印当前占用 apt 的进程，便于排查
+_apt_lock_holders() {
+  ps -eo pid,cmd 2>/dev/null | grep -E '[u]nattended-upgr|[a]pt-get|[a]pt |[d]pkg ' | head -n 8 || true
+  if command -v fuser >/dev/null 2>&1; then
+    fuser -v /var/lib/dpkg/lock-frontend /var/lib/dpkg/lock /var/lib/apt/lists/lock 2>&1 | head -n 12 || true
+  fi
+}
+
+# 部署期间临时停自动升级（重启后常立刻再跑，占锁很久）
+_stop_auto_apt_for_deploy() {
+  if [[ "${APT_STOP_AUTO:-1}" != "1" ]]; then
+    return 0
+  fi
+  if command -v systemctl >/dev/null 2>&1; then
+    log "临时停止 unattended-upgrades / apt-daily，避免占锁…"
+    systemctl stop unattended-upgrades.service 2>/dev/null || true
+    systemctl stop apt-daily.service apt-daily-upgrade.service 2>/dev/null || true
+    systemctl kill --kill-who=all unattended-upgrades.service 2>/dev/null || true
+  fi
+  # 若仍卡死且 APT_FORCE_UNLOCK=1，才强杀（默认不杀，避免打断半截升级）
+  if [[ "${APT_FORCE_UNLOCK:-0}" == "1" ]]; then
+    log "APT_FORCE_UNLOCK=1：结束残留 apt/dpkg 进程…"
+    pkill -9 -x unattended-upgr 2>/dev/null || true
+    pkill -9 -x apt-get 2>/dev/null || true
+    pkill -9 -x apt 2>/dev/null || true
+    pkill -9 -x dpkg 2>/dev/null || true
+    sleep 2
+    dpkg --configure -a 2>/dev/null || true
+  fi
+}
+
 # 等待 apt/dpkg 锁（常见：unattended-upgrades 正在跑）
 wait_for_apt_lock() {
   local max_wait="${1:-600}"
@@ -41,6 +72,7 @@ wait_for_apt_lock() {
   local lock_frontend="/var/lib/dpkg/lock-frontend"
   local lock_dpkg="/var/lib/dpkg/lock"
   local lock_lists="/var/lib/apt/lists/lock"
+  local tried_stop=0
 
   if ! command -v apt-get >/dev/null 2>&1; then
     return 0
@@ -69,13 +101,22 @@ wait_for_apt_lock() {
       return 0
     fi
     if [[ "${waited}" -eq 0 ]]; then
-      log "检测到 apt/dpkg 正被占用（多为 unattended-upgrades），等待释放…"
-      log "可另开终端查看: ps aux | grep -E 'unattended|apt|dpkg'"
+      log "检测到 apt/dpkg 正被占用（多为重启后的 unattended-upgrades），等待释放…"
+      log "占用进程："
+      _apt_lock_holders | while IFS= read -r line; do log "  ${line}"; done
+      log "可另开终端: ps aux | grep -E 'unattended|apt|dpkg'"
+    elif [[ "${tried_stop}" -eq 0 && "${waited}" -ge 15 ]]; then
+      # 等约 15s 仍占用：停自动升级服务（不杀半截 dpkg，除非 APT_FORCE_UNLOCK=1）
+      tried_stop=1
+      _stop_auto_apt_for_deploy
     elif [[ $((waited % 60)) -eq 0 ]]; then
       log "仍在等待 apt 锁…已等 ${waited}s / ${max_wait}s"
+      _apt_lock_holders | while IFS= read -r line; do log "  ${line}"; done
     fi
     if [[ "${waited}" -ge "${max_wait}" ]]; then
-      die "等待 apt 锁超时(${max_wait}s)。请稍后再跑一键部署，或手动: kill 结束后再执行 manage.sh"
+      log "超时前占用进程："
+      _apt_lock_holders | while IFS= read -r line; do log "  ${line}"; done
+      die "等待 apt 锁超时(${max_wait}s)。可手动: systemctl stop unattended-upgrades; 或 APT_FORCE_UNLOCK=1 bash manage.sh"
     fi
     sleep 5
     waited=$((waited + 5))