Files
HomeAgent/deploy-plan.sh
JianFeeeee 55906b8b7e chore(deploy): 补强部署后验证
原来只说「等 20s 看 /kernel 状态」,等于没验。进程活着 ≠ agent 起来了。
现在逐项核对,每项对应一个真实故障模式:

- 60s 内未见 'kernel ready' ⇒ 判失败并给出回滚命令 + journalctl 尾部
- 注册工具条目数(生产应 18 个左右)⇒ 少了说明插件加载异常
- 'LLM API unreachable' 次数 > 3 ⇒ 内核在 rollback 循环
  (生产设了 max_retries=100000,不可达会一直重试)
- ONNX provider 相关日志 ⇒ 缺失时应是明确错误+降级,不静默假装启用
2026-09-27 19:12:12 +08:00

172 lines
6.2 KiB
Bash
Executable File
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#!/bin/bash
# 生产部署方案(**待确认,不自动执行**)
#
# 用法:
# bash deploy-plan.sh check # 只做部署前检查,不改任何东西
# bash deploy-plan.sh backup # 备份当前二进制与适配器
# bash deploy-plan.sh deploy # 替换二进制并重启(需显式确认)
# bash deploy-plan.sh rollback # 回滚到备份
#
# ★ 本脚本刻意**不包含**任何"自动回滚"逻辑:回滚要不要做、什么时候做,
# 是人的判断。脚本只负责把状态保全好,让回滚成为一条可执行的命令。
set -uo pipefail
BIN=/usr/local/bin/homed
DATA=/home/newqqagent
BAK=/var/tmp/homed-backup-$(date +%Y%m%d-%H%M%S)
NEW=${NEW_BIN:-/tmp/homed-ort}
ADAPTERS=$DATA/adapters
SVC=homeagent.service
say() { printf '\n\033[1m%s\033[0m\n' "$*"; }
info() { printf ' %s\n' "$*"; }
# ---------------------------------------------------------------- check
do_check() {
say "部署前检查"
info "服务状态: $(systemctl is-active $SVC)"
info "当前二进制: $BIN ($(stat -c %s $BIN) 字节, $(stat -c %y $BIN | cut -d. -f1))"
if [ ! -x "$NEW" ]; then
info "★ 新二进制不存在: $NEW"
return 1
fi
info "新二进制: $NEW ($(stat -c %s $NEW) 字节)"
# ★ 最关键的一条:必须是 onnxruntime 构建,否则依存句法/多模态失效
if ! go version -m "$NEW" 2>/dev/null | grep -Eq 'build[[:space:]]+-tags=.*onnxruntime'; then
info "★★ 新二进制**不是** onnxruntime 构建 —— 拒绝部署"
info " (package-linux.sh:139 同样会拒绝;缺它会让依存句法与多模态失效)"
return 1
fi
info "✓ onnxruntime 构建标签正确"
# 运行期依赖
if [ ! -f /opt/onnxruntime/libonnxruntime.so ] \
&& ! ls /usr/local/lib/libonnxruntime.so* /usr/lib/libonnxruntime.so* >/dev/null 2>&1; then
info "★ 找不到 libonnxruntime.so —— provider 会降级"
else
info "✓ libonnxruntime.so 就位"
fi
# 模型资产(运行期目录,不影响编译)
local mdl=$(du -sh $DATA/models 2>/dev/null | awk '{print $1}')
info "模型资产: ${mdl:-无}(运行期目录,部署不改动)"
# 适配器:d3eaff4 的新逻辑在"无历史清单"时不动文件
if [ -f "$ADAPTERS/.bundled" ]; then
info "适配器清单: 存在(升级时落后的会被更新,用户改过的会保留)"
else
info "适配器清单: 无 ⇒ 首次升级**不动**任何已存在的适配器文件"
fi
info "适配器文件数: $(ls $ADAPTERS/*.lua 2>/dev/null | wc -l)"
return 0
}
# ---------------------------------------------------------------- backup
do_backup() {
say "备份"
mkdir -p "$BAK"
cp -a "$BIN" "$BAK/homed.bin"
[ -d "$ADAPTERS" ] && cp -a "$ADAPTERS" "$BAK/adapters"
cp -a /etc/systemd/system/$SVC "$BAK/" 2>/dev/null || true
info "已备份到: $BAK"
info " homed.bin / adapters / unit 文件"
# 回滚命令
cat > "$BAK/ROLLBACK.sh" <<RB
#!/bin/bash
# 回滚到 $BAK
set -e
systemctl stop $SVC
cp -a $BAK/homed.bin $BIN
[ -d $BAK/adapters ] && rm -rf $ADAPTERS && cp -a $BAK/adapters $ADAPTERS
systemctl start $SVC
systemctl is-active $SVC
RB
chmod +x "$BAK/ROLLBACK.sh"
info "回滚命令已生成: bash $BAK/ROLLBACK.sh"
}
# ---------------------------------------------------------------- deploy
do_deploy() {
say "部署"
do_check || { info "检查未通过,中止"; return 1; }
echo
read -r -p "确认替换 $BIN 并重启 $SVC ? 输入 yes 继续: " ans
[ "$ans" = "yes" ] || { info "已取消"; return 1; }
do_backup
say "替换二进制"
systemctl stop "$SVC"
# ★ install 而非 cp:保留 setuid/权限语义,且原子替换
install -m 0755 "$NEW" "$BIN"
info "已安装: $BIN ($(stat -c %s $BIN) 字节)"
systemctl start "$SVC"
sleep 8
if systemctl is-active --quiet "$SVC"; then
info "✓ 服务已启动: $(systemctl is-active $SVC)"
else
info "✗ 服务启动失败 —— 回滚:bash $BAK/ROLLBACK.sh"
journalctl -u "$SVC" --since '-2 min' --no-pager | tail -20
return 1
fi
say "部署后验证"
# 只说"等 20s 看日志"是不够的 —— 进程活着 ≠ agent 起来了。
# 逐项核对,每项都对应一个真实故障模式:
info "等待 agent 就绪(最多 60s)…"
local ready=0
for i in $(seq 1 30); do
if journalctl -u "$SVC" --since '-2 min' --no-pager 2>/dev/null \
| grep -q 'kernel ready'; then ready=1; break; fi
sleep 2
done
if [ "$ready" = "1" ]; then
info "✓ kernel ready"
else
info "✗ 60s 内没看到 'kernel ready' —— 回滚:bash $BAK/ROLLBACK.sh"
journalctl -u "$SVC" --since '-2 min' --no-pager | tail -25
return 1
fi
# 插件加载:生产有 18 个,少于 10 个说明加载异常
local nplug
nplug=$(journalctl -u "$SVC" --since '-2 min' --no-pager 2>/dev/null \
| grep -c 'SetToolRegistrar registering tool')
info "注册工具条目: $nplug"
# LLM 可达:不可达会进 rollback 循环(生产设了 max_retries=100000)
local unreach
unreach=$(journalctl -u "$SVC" --since '-2 min' --no-pager 2>/dev/null \
| grep -c 'LLM API unreachable')
if [ "${unreach:-0}" -gt 3 ]; then
info "✗ LLM 不可达 ${unreach} 次 —— 内核在 rollback 循环"
info " 回滚:bash $BAK/ROLLBACK.sh"
return 1
fi
info "✓ LLM 可达(unreachable=${unreach:-0})"
info "✓ ONNX provider:见日志 'onnx' 相关行(缺失时应为明确错误+降级,不静默)"
journalctl -u "$SVC" --since '-2 min' --no-pager | grep -iE 'onnx|provider' | tail -5
}
# ---------------------------------------------------------------- rollback
do_rollback() {
say "回滚"
local latest
latest=$(ls -dt /var/tmp/homed-backup-* 2>/dev/null | head -1)
if [ -z "$latest" ]; then info "找不到备份"; return 1; fi
info "使用备份: $latest"
systemctl stop "$SVC"
cp -a "$latest/homed.bin" "$BIN"
[ -d "$latest/adapters" ] && { rm -rf "$ADAPTERS"; cp -a "$latest/adapters" "$ADAPTERS"; }
systemctl start "$SVC"
sleep 5
info "回滚后: $(systemctl is-active $SVC)"
}
case "${1:-check}" in
check) do_check ;;
backup) do_backup ;;
deploy) do_deploy ;;
rollback) do_rollback ;;
*) echo "用法: $0 {check|backup|deploy|rollback}"; exit 2 ;;
esac