#!/usr/bin/env bash
# 织忆 MemoryWeave 统一部署脚本 — build two + install two + restart two + verify
# 2026-10-08 · t_f57b8bbf · 根治「重启 zhiyid 不重启 sidecar → Rust 改部署后 sidecar 跑旧二进制静默故障」事故
#
# 用法:
#   ./scripts/deploy.sh            完整部署: 构建两+装两+停两+启两+校验
#   ./scripts/deploy.sh --verify-only   只跑校验（不构建/不安装/不重启）—— 反例测试/人工体检用
#   make deploy                    = ./scripts/deploy.sh
set -euo pipefail

REPO_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
BIN_DIR="${BIN_DIR:-/home/muc/bin}"

# 构建产物（持久路径，均已 gitignore）
BUILD_GO="$REPO_DIR/go/cmd/zhiyid/zhiyid"
BUILD_RUST="$REPO_DIR/rust/target/release/zhiyi-consolidate"
# 安装落点（与 ~/.config/systemd/user/*.service 的 ExecStart 一致）
INSTALL_GO="$BIN_DIR/zhiyid-new"
INSTALL_RUST="$BIN_DIR/zhiyi-consolidate"
# 服务名（用户级 systemd）
SVC_GO="zhiyid.service"
SVC_RUST="zhiyi-consolidate.service"
HEALTH_URL="${HEALTH_URL:-http://127.0.0.1:7821/health}"

log() { printf '\033[1;34m[deploy]\033[0m %s\n' "$*"; }
die() { printf '\033[1;31m[deploy:FAIL]\033[0m %s\n' "$*" >&2; exit 1; }

# 校验单个服务的运行态是否=本次构建的二进制
# 判据优先级（纪律）: 副作用(PID/inode/校验和) > 返回值 > 自报
verify_service() {
  local name="$1" build_path="$2" pid exe run_sha build_sha
  # ① is-active
  [ "$(systemctl --user is-active "$name")" = "active" ] || die "$name 非 active"
  # ② 取 MainPID
  pid="$(systemctl --user show -p MainPID --value "$name")"
  [ -n "$pid" ] && [ "$pid" != "0" ] || die "$name: MainPID 为空/0"
  # ③ 进程 exe 不得指向已删除 inode（旧二进制被 install 覆盖后、进程仍跑旧 inode → readlink 含 "(deleted)"）
  exe="$(readlink "/proc/$pid/exe" 2>/dev/null || true)"
  case "$exe" in
    "") die "$name: 无法 readlink /proc/$pid/exe" ;;
    *" (deleted)"*) die "$name: 进程仍跑已删除 inode → 二进制被换过但进程没重启 ($exe)" ;;
  esac
  # ④ 运行中二进制校验和 == 本次构建产物（抓「磁盘装了新版但进程仍跑旧版」静默故障）
  run_sha="$(sha256sum "/proc/$pid/exe" 2>/dev/null | awk '{print $1}')"
  build_sha="$(sha256sum "$build_path" 2>/dev/null | awk '{print $1}')"
  [ -n "$run_sha" ] || die "$name: 无法读取运行中二进制"
  [ "$run_sha" = "$build_sha" ] || die "$name: 运行中二进制 ≠ 本次构建产物 (运行=$run_sha 构建=$build_sha) → sidecar 静默跑旧版"
  log "✔ $name active · PID=$pid · exe=$exe · 校验和匹配本次构建"
}

# 深健康检查: /health 连 sidecar 可达性（P0 已做，故障返 503）→ 期望 200
verify_health() {
  local code
  code="$(curl -s -o /dev/null -w '%{http_code}' -m 5 "$HEALTH_URL")"
  [ "$code" = "200" ] || die "/health 返回 $code (期望 200) → zhiyid 或 sidecar 故障"
  log "✔ /health = $code"
}

do_build() {
  log "构建 Go → $BUILD_GO"
  ( cd "$REPO_DIR/go" && CGO_ENABLED=1 go build -o cmd/zhiyid/zhiyid ./cmd/zhiyid )
  log "构建 Rust → $BUILD_RUST"
  ( cd "$REPO_DIR/rust" && cargo build --release )
}

do_install() {
  log "停止 $SVC_GO + $SVC_RUST"
  systemctl --user stop "$SVC_GO" || true
  systemctl --user stop "$SVC_RUST" || true
  # 用 install 而非 cp：install 先写临时再 rename，避免 Text file busy（进程占着旧 inode 时 cp 会 EBUSY）
  log "install 二进制（不用 cp，避免 Text file busy）"
  install -m 0755 "$BUILD_GO" "$INSTALL_GO"
  install -m 0755 "$BUILD_RUST" "$INSTALL_RUST"
  log "daemon-reload"
  systemctl --user daemon-reload
  log "启动 $SVC_GO + $SVC_RUST"
  systemctl --user start "$SVC_GO"
  systemctl --user start "$SVC_RUST"
  sleep 2
}

do_verify() {
  log "校验 ..."
  verify_service "$SVC_GO" "$BUILD_GO"
  verify_service "$SVC_RUST" "$BUILD_RUST"
  verify_health
  log "✅ 部署成功: 两服务 active · 进程无 (deleted) · 运行二进制=本次构建 · /health=200"
}

if [ "${1:-}" = "--verify-only" ]; then
  do_verify
  exit 0
fi

do_build
do_install
do_verify