pj0235-eai_agentplatform/start_dev_10231_10232.sh

#!/usr/bin/env bash
if [ -z "${BASH_VERSION:-}" ]; then
  if command -v bash >/dev/null 2>&1; then
    echo "[INFO] start_dev_10231_10232.sh requires bash. Re-running with bash ..."
    exec bash "$0" "$@"
  fi
  echo "[FATAL] bash is required but was not found on PATH." >&2
  echo "        Please run: bash ./start_dev_10231_10232.sh" >&2
  exit 1
fi

set -euo pipefail

FORCE=0
SKIP_PREFLIGHT=0
SKIP_SMOKE=0
SKIP_BUILD=0

while [[ $# -gt 0 ]]; do
  case "$1" in
    --force|-f)
      FORCE=1
      ;;
    --skip-preflight)
      SKIP_PREFLIGHT=1
      ;;
    --skip-smoke)
      SKIP_SMOKE=1
      ;;
    --skip-build)
      SKIP_BUILD=1
      ;;
    --help|-h)
      cat <<'EOF'
Usage: ./start_dev_10231_10232.sh [options]

Options:
  --force            Kill whatever holds the ports, even non-project processes
                     (a previous run of this script is always stopped and
                     replaced automatically)
  --skip-preflight   Skip preflight checks
  --skip-smoke       Skip admin account smoke login
  --skip-build       Do not rebuild the backend, even if sources are newer
  --help             Show this help
EOF
      exit 0
      ;;
    *)
      echo "[FATAL] Unknown option: $1" >&2
      exit 1
      ;;
  esac
  shift
done

# ================================================================
# 端口说明
#   10231  - 前端 Vite 开发服务器 (dev only)
#   10232  - 后端 Go API 服务 (dev only)
#   8080   - 外部 LLM 服务 (llamacpp) — AI 对话 API 入口
#   8081   - 外部 Embed 服务 (llamacpp) — 向量嵌入服务
#   11434  - 外部 Ollama 服务 (如使用 Ollama 路由)
# ================================================================
BACKEND_PORT=10232
FRONTEND_PORT=10231

REPO_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
APP_DIR="$REPO_DIR/eai_agentplatform"
BACKEND_DIR="$APP_DIR/backend-go"
BACKEND_BIN="$BACKEND_DIR/bin/eai_agentplatform-server"
FRONTEND_DIR="$APP_DIR/frontend"
DEV_BIND_HOST="${EAI_DEV_BIND_HOST:-0.0.0.0}"
DEBUG_LOG_DIR="$REPO_DIR/debuglog"
mkdir -p "$DEBUG_LOG_DIR"

RUN_ID="$(date +%Y%m%d_%H%M%S)"
BACKEND_LOG="$DEBUG_LOG_DIR/backend_${RUN_ID}.log"
FRONTEND_LOG="$DEBUG_LOG_DIR/frontend_${RUN_ID}.log"
BACKEND_PID_FILE="$DEBUG_LOG_DIR/backend.pid"
FRONTEND_PID_FILE="$DEBUG_LOG_DIR/frontend.pid"

# 加载本地运行 secrets（.env 已被 .gitignore 忽略，不进版本库）。
# 例如 SERPAPI_API_KEY 等环境变量，供后端进程继承（backed 用 `env PORT=... $BIN` 启动，
# 会继承本 shell 导出的所有环境变量）。文件不存在时静默跳过。
if [ -f "$APP_DIR/.env" ]; then
  set -a
  # shellcheck disable=SC1091
  source "$APP_DIR/.env"
  set +a
fi

check() {
  local label="$1"
  local fix_hint="$2"
  shift 2
  printf '  [.] %s ...' "$label"
  if "$@" >/dev/null 2>&1; then
    printf ' OK\n'
    return 0
  fi
  printf ' FAIL\n'
  if [[ -n "$fix_hint" ]]; then
    printf '      fix: %s\n' "$fix_hint"
  fi
  return 1
}

wait_health() {
  local url="$1"
  local label="$2"
  local timeout_sec="${3:-60}"
  local elapsed=0
  printf '  [.] waiting on %s (%s) ...' "$label" "$url"
  while (( elapsed < timeout_sec )); do
    if curl -fsS --max-time 2 "$url" >/dev/null 2>&1; then
      printf ' READY\n'
      return 0
    fi
    sleep 1
    elapsed=$((elapsed + 1))
    printf '.'
  done
  printf ' TIMEOUT\n'
  return 1
}

detect_public_host() {
  local candidate
  for candidate in $(hostname -I 2>/dev/null || true); do
    case "$candidate" in
      127.*|172.17.*)
        continue
        ;;
      *)
        echo "$candidate"
        return 0
        ;;
    esac
  done
  echo "127.0.0.1"
}

port_pids() {
  local port="$1"
  if command -v lsof >/dev/null 2>&1; then
    lsof -tiTCP:"$port" -sTCP:LISTEN 2>/dev/null | sort -u
    return 0
  fi
  if command -v ss >/dev/null 2>&1; then
    ss -ltnp "( sport = :$port )" 2>/dev/null | awk -F'pid=' 'NR>1 && NF>1 {split($2,a,","); print a[1]}' | sort -u
    return 0
  fi
  return 1
}

# Decide whether a process belongs to this project.
#
# This must look at the full command line, not `ps -o comm=`: comm is capped at
# 15 characters (the kernel's TASK_COMM_LEN), so the backend reports itself as
# "eai_agentplatfo" and could never match its own name. That made the launcher
# treat its own backend as a foreign process and refuse to restart.
process_is_ours() {
  local pid="$1"
  local port="$2"
  local args
  args="$(ps -o args= -p "$pid" 2>/dev/null || true)"
  [[ -z "$args" ]] && return 1
  [[ "$args" == *"$BACKEND_BIN"* ]] && return 0
  [[ "$args" == *vite* && "$args" == *"--port $port"* ]] && return 0
  return 1
}

process_label() {
  local pid="$1"
  local args
  args="$(ps -o args= -p "$pid" 2>/dev/null || true)"
  if [[ -z "$args" ]]; then
    printf 'unknown'
    return 0
  fi
  args="${args%% *}"
  printf '%s' "${args##*/}"
}

# Stop the run recorded by a previous invocation. Uses the pid files so a
# previous instance still gets cleaned up even if it is not holding its port.
kill_previous_run() {
  local pid_file="$1"
  local port="$2"
  local label="$3"
  local pid
  [[ -f "$pid_file" ]] || return 0
  pid="$(cat "$pid_file" 2>/dev/null || true)"
  [[ "$pid" =~ ^[0-9]+$ ]] || return 0
  kill -0 "$pid" 2>/dev/null || return 0
  process_is_ours "$pid" "$port" || return 0
  printf '  [.] stopping previous %s (PID %s)\n' "$label" "$pid"
  kill "$pid" 2>/dev/null || true
  sleep 1
  kill -9 "$pid" 2>/dev/null || true
  return 0
}

resolve_port() {
  local port="$1"
  local label="$2"
  local pids
  mapfile -t pids < <(port_pids "$port" || true)
  if [[ ${#pids[@]} -eq 0 ]]; then
    printf '  [.] %s :%s free\n' "$label" "$port"
    return 0
  fi

  printf '  [!] %s :%s in use by ' "$label" "$port"
  for pid in "${pids[@]}"; do
    printf '%s(PID %s) ' "$(process_label "$pid")" "$pid"
  done
  printf '\n'

  for pid in "${pids[@]}"; do
    if [[ "$FORCE" -eq 1 ]] || process_is_ours "$pid" "$port"; then
      printf '      killing previous %s(PID %s)\n' "$(process_label "$pid")" "$pid"
      kill "$pid" 2>/dev/null || true
      sleep 1
      kill -9 "$pid" 2>/dev/null || true
    else
      printf '      keeping non-project process: %s(PID %s)\n' "$(process_label "$pid")" "$pid"
      printf '      fix: stop that process manually, or rerun with --force\n'
      return 1
    fi
  done

  mapfile -t pids < <(port_pids "$port" || true)
  if [[ ${#pids[@]} -gt 0 ]]; then
    printf '  [FAIL] %s :%s still in use\n' "$label" "$port"
    return 1
  fi

  printf '  [OK] %s :%s released\n' "$label" "$port"
  return 0
}

# True when the backend binary is missing or older than its sources.
#
# Preflight only proves the binary exists, which happily launches stale code
# after a source edit — the failure is silent, so it needs an explicit check.
backend_needs_build() {
  [[ -f "$BACKEND_BIN" ]] || return 0
  local newer
  newer="$(find "$BACKEND_DIR" -name '*.go' -newer "$BACKEND_BIN" -print -quit 2>/dev/null || true)"
  [[ -n "$newer" ]] && return 0
  local f
  for f in "$BACKEND_DIR/go.mod" "$BACKEND_DIR/go.sum"; do
    [[ -f "$f" && "$f" -nt "$BACKEND_BIN" ]] && return 0
  done
  return 1
}

build_backend() {
  local build_log="$DEBUG_LOG_DIR/backend_build.log"
  if ! command -v go >/dev/null 2>&1; then
    printf '  [FAIL] backend sources changed, but go is not on PATH\n'
    printf '      fix: install go, or rebuild manually then rerun with --skip-build\n'
    return 1
  fi
  printf '  [.] backend sources newer than binary, rebuilding ...'
  if ( cd "$BACKEND_DIR" && go build -o bin/eai_agentplatform-server ./cmd/server ) >"$build_log" 2>&1; then
    printf ' OK\n'
    return 0
  fi
  printf ' FAIL\n'
  printf '      log : %s\n' "$build_log"
  tail -n 20 "$build_log" 2>/dev/null | sed 's/^/      /' || true
  return 1
}

cleanup_started() {
  if [[ -f "$BACKEND_PID_FILE" ]]; then
    kill "$(cat "$BACKEND_PID_FILE")" 2>/dev/null || true
  fi
  if [[ -f "$FRONTEND_PID_FILE" ]]; then
    kill "$(cat "$FRONTEND_PID_FILE")" 2>/dev/null || true
  fi
}

echo "==> eai_agentplatform dev launcher"
echo "    repo     : $REPO_DIR"
echo "    backend  : $BACKEND_DIR (port $BACKEND_PORT)"
echo "    frontend : $FRONTEND_DIR (port $FRONTEND_PORT)"
echo "    bind host: $DEV_BIND_HOST"
echo

if [[ "${EAI_DEV_PUBLIC_HOST:-}" != "" ]]; then
  DEV_PUBLIC_HOST="$EAI_DEV_PUBLIC_HOST"
elif [[ "$DEV_BIND_HOST" == "0.0.0.0" ]]; then
  DEV_PUBLIC_HOST="$(detect_public_host)"
else
  DEV_PUBLIC_HOST="$DEV_BIND_HOST"
fi

if [[ "$SKIP_PREFLIGHT" -eq 0 ]]; then
  echo "[1/4] preflight checks"
  all_ok=1
  check "backend dir exists"      "run from repo root"               test -d "$BACKEND_DIR"                || all_ok=0
  check "frontend dir exists"     "run from repo root"               test -d "$FRONTEND_DIR"               || all_ok=0
  check "frontend node_modules"   "run npm install in frontend/"     test -e "$FRONTEND_DIR/node_modules"  || all_ok=0
  check "node on PATH"            "install nodejs"                   command -v node                       || all_ok=0
  check "npm on PATH"             "install npm"                      command -v npm                        || all_ok=0
  check "curl on PATH"            "install curl"                     command -v curl                       || all_ok=0
  if [[ "$all_ok" -eq 0 ]]; then
    echo
    echo "[ABORT] preflight failed."
    exit 1
  fi
  echo
else
  echo "[1/4] preflight skipped (--skip-preflight)"
  echo
fi

# Runs whether or not preflight did: a stale binary is a silent failure, and
# building before the old services are stopped means a compile error leaves the
# currently running stack untouched.
if [[ "$SKIP_BUILD" -eq 1 ]]; then
  echo "[1/4] backend build check skipped (--skip-build)"
  echo
elif backend_needs_build; then
  if ! build_backend; then
    echo
    echo "[ABORT] backend build failed; services left as they were."
    exit 1
  fi
  echo
else
  echo "[1/4] backend binary up to date"
  echo
fi

echo "[2/4] port conflict check"
kill_previous_run "$BACKEND_PID_FILE" "$BACKEND_PORT" "backend"
kill_previous_run "$FRONTEND_PID_FILE" "$FRONTEND_PORT" "frontend"
resolve_port "$BACKEND_PORT" "backend" || exit 1
resolve_port "$FRONTEND_PORT" "frontend" || exit 1
echo

trap cleanup_started EXIT

echo "[3/4] launch services"
echo "  [.] starting backend (eai_agentplatform-server :$BACKEND_PORT)"
echo "  [.] log : $BACKEND_LOG"
nohup env PORT="$BACKEND_PORT" "$BACKEND_BIN" >>"$BACKEND_LOG" 2>&1 </dev/null &
BACKEND_PID=$!
echo "$BACKEND_PID" > "$BACKEND_PID_FILE"

if ! wait_health "http://127.0.0.1:$BACKEND_PORT/api/health" "backend"; then
  echo
  echo "[ABORT] backend did not become healthy, frontend not started."
  echo "        check: $BACKEND_LOG"
  exit 1
fi

echo "  [.] starting frontend (vite :$FRONTEND_PORT)"
echo "  [.] log : $FRONTEND_LOG"
nohup bash -lc "cd \"$FRONTEND_DIR\" && exec node ./node_modules/vite/bin/vite.js --host \"$DEV_BIND_HOST\" --port \"$FRONTEND_PORT\"" >>"$FRONTEND_LOG" 2>&1 </dev/null &
FRONTEND_PID=$!
echo "$FRONTEND_PID" > "$FRONTEND_PID_FILE"
echo

echo "[4/4] health check + smoke login"
f_ok=1
if ! wait_health "http://127.0.0.1:$FRONTEND_PORT/" "frontend"; then
  f_ok=0
fi

if [[ "$SKIP_SMOKE" -eq 0 && "$f_ok" -eq 1 ]]; then
  echo
  echo "  [.] smoke login admin account ..."
  payload='{"username":"admin","password":"admin123"}'
  status_code="$(curl -sS -o /dev/null -w '%{http_code}' \
    -H 'Content-Type: application/json' \
    -d "$payload" \
    "http://127.0.0.1:$BACKEND_PORT/api/auth/login" || true)"
  if [[ "$status_code" == "200" ]]; then
    printf '      [OK]   admin login success\n'
  else
    printf '      [FAIL] admin login (HTTP %s)\n' "$status_code"
  fi
else
  echo "  [.] smoke login skipped"
fi

echo
echo "================================================================="
echo "  Service status"
echo "================================================================="
echo "    backend  : http://$DEV_PUBLIC_HOST:$BACKEND_PORT"
echo "    frontend : http://$DEV_PUBLIC_HOST:$FRONTEND_PORT"
echo "    backend log : $BACKEND_LOG"
echo "    frontend log: $FRONTEND_LOG"
echo "    PIDs: backend=$BACKEND_PID frontend=$FRONTEND_PID"
echo "================================================================="
echo

trap - EXIT
echo "Usage: ./start_dev_10231_10232.sh [--force] [--skip-preflight] [--skip-build]"
echo "       env: EAI_DEV_BIND_HOST=0.0.0.0 (listen all)"
echo "       env: EAI_DEV_PUBLIC_HOST=x.x.x.x (public url)"
echo