feat: run mnote-web prod under systemd --user with auto-restart

Replace detached spawn with mnote-web-prod.service (Restart=always, enable + Linger). prod:start installs the unit, writes env/current symlink, and restarts via systemctl so the gateway survives kills and logout. Also drop unused std::fs import in web_shell.
2026-07-25 14:56:42 +08:00
parent 262e66b02e
commit 2deaf59f7b
3 changed files with 279 additions and 28 deletions
@@ -9,6 +9,10 @@
 * - 新 release 从 0.0.1 开始递增，0.0.9 后是 0.1.0。
 * - 每个 release 保留独立源码快照和二进制，不覆盖旧 release。
 * - 默认启动端口是 3003，避免占用开发端口 3000。
+ * - 进程由 systemd --user（mnote-web-prod.service）托管：Restart=always + enable，
+ *   本机需 loginctl Linger=yes（已配置则登出后仍运行）。
+ * - 管理：systemctl --user status|restart|stop mnote-web-prod.service
+ *   日志：journalctl --user -u mnote-web-prod.service -f
 */

 const fs = require("node:fs");
@@ -23,11 +27,47 @@ const RELEASE_ROOT = path.join(ROOT, "dist", "releases", "mnote-web");
 const BUILD_CACHE_DIR = path.join(ROOT, "dist", "prod-build-cache", "mnote-web");
 const RUN_DIR = path.join(ROOT, "dist", "run", "mnote-web-prod");
 const RUN_STATE_PATH = path.join(RUN_DIR, "process.json");
+const RUN_ENV_PATH = path.join(RUN_DIR, "env");
+const RUN_CURRENT_LINK = path.join(RUN_DIR, "current");
+const SYSTEMD_UNIT_NAME = "mnote-web-prod.service";
+const SYSTEMD_UNIT_SOURCE = path.join(ROOT, "deploy", "systemd", "mnote-web-prod.service");
+const SYSTEMD_USER_DIR = path.join(os.homedir(), ".config", "systemd", "user");
+const SYSTEMD_UNIT_TARGET = path.join(SYSTEMD_USER_DIR, SYSTEMD_UNIT_NAME);
 const DEFAULT_PORT = 3003;
 const DEFAULT_TURSO_LOCAL_PATH = "/mnt/Data1T/Mnote_data/control-plane/control-plane-prod-libsql.db";
 const DEFAULT_CONTROL_PLANE_BACKEND = "libsql-local";
 const ENV_ALL_PATH = path.join(ROOT, ".env.all");
 const RELEASE_BUILD_SCHEMA = "mnote-web-prod-runtime-assets-v2";
+// 不写入 systemd EnvironmentFile 的敏感/会话壳变量
+const ENV_FILE_DENYLIST = new Set([
+  "DBUS_SESSION_BUS_ADDRESS",
+  "DISPLAY",
+  "SSH_AUTH_SOCK",
+  "SSH_CLIENT",
+  "SSH_CONNECTION",
+  "SSH_TTY",
+  "XAUTHORITY",
+  "WAYLAND_DISPLAY",
+  "GPG_AGENT_INFO",
+  "OLDPWD",
+  "PWD",
+  "SHLVL",
+  "_",
+  "npm_config_user_agent",
+  "npm_lifecycle_event",
+  "npm_lifecycle_script",
+  "npm_package_name",
+  "npm_package_version",
+  "npm_command",
+  "npm_execpath",
+  "npm_node_execpath",
+  "INIT_CWD",
+  "COLOR",
+  "COLORTERM",
+  "TERM",
+  "TERM_PROGRAM",
+  "TERM_PROGRAM_VERSION",
+]);
 const SUPPLEMENTAL_RUNTIME_ASSETS = [
  {
    source: "node_modules/jszip/dist/jszip.min.js",
@@ -516,6 +556,174 @@ async function verifyStartupAssets(port) {
  }
 }

+function escapeEnvFileValue(value) {
+  const text = String(value ?? "");
+  // systemd EnvironmentFile：含空白/引号/反斜杠时用双引号并转义
+  if (/[\s"'\\$`]/.test(text) || text === "") {
+    return `"${text.replace(/\\/g, "\\\\").replace(/"/g, '\\"').replace(/\$/g, "\\$")}"`;
+  }
+  return text;
+}
+
+function buildProdRuntimeEnv(port, bind, publicBind, controlPlaneBackend) {
+  // 只落 .env.all + 生产覆盖，不把整份 shell/npm 环境写进 systemd
+  const env = {
+    ...loadEnvFile(ENV_ALL_PATH),
+  };
+  // 允许调用方在进程环境里临时覆盖关键项（如 MNOTE_PROD_PORT）
+  const passThroughKeys = [
+    "MNOTE_PROD_PORT",
+    "MNOTE_WEB_BIND",
+    "MNOTE_WEB_PUBLIC_BIND",
+    "MNOTE_CONTROL_PLANE_BACKEND",
+    "MNOTE_TURSO_LOCAL_PATH",
+    "MNOTE_TURSO_URL",
+    "MNOTE_TURSO_AUTH_TOKEN",
+    "MNOTE_TURSO_SYNC_URL",
+    "RUST_LOG",
+    "MNOTE_LIGHTRAG_API_KEY",
+    "MNOTE_LIGHTRAG_BASE_URL",
+    "LIGHTRAG_URL",
+    "LIGHTRAG_API_KEY",
+  ];
+  for (const key of passThroughKeys) {
+    if (process.env[key] !== undefined && process.env[key] !== "") {
+      env[key] = process.env[key];
+    }
+  }
+
+  env.MNOTE_WEB_BIND = bind;
+  env.MNOTE_WEB_PUBLIC_BIND = publicBind;
+  env.MNOTE_CONTROL_PLANE_BACKEND = controlPlaneBackend;
+  env.RUST_LOG = env.RUST_LOG || process.env.RUST_LOG || "info";
+  if (
+    controlPlaneBackend === "libsql-local"
+    || controlPlaneBackend === "turso-local"
+    || controlPlaneBackend === "turso"
+  ) {
+    env.MNOTE_TURSO_LOCAL_PATH =
+      env.MNOTE_TURSO_LOCAL_PATH
+      || process.env.MNOTE_TURSO_LOCAL_PATH
+      || DEFAULT_TURSO_LOCAL_PATH;
+  }
+  env.MNOTE_PROD_PORT = String(port);
+  // PATH 给子进程一个干净可预期值（不依赖 npm 注入的 node_modules/.bin）
+  env.PATH = [
+    path.join(os.homedir(), ".local", "bin"),
+    path.join(os.homedir(), ".cargo", "bin"),
+    path.join(os.homedir(), ".npm-global", "bin"),
+    "/usr/local/bin",
+    "/usr/bin",
+    "/bin",
+  ].join(":");
+  env.HOME = os.homedir();
+  env.LANG = process.env.LANG || "C.UTF-8";
+  return env;
+}
+
+async function writeSystemdEnvFile(env) {
+  const lines = [
+    "# Generated by scripts/prod-build-start.js — do not edit by hand",
+    `# generatedAt=${new Date().toISOString()}`,
+  ];
+  const keys = Object.keys(env).sort();
+  for (const key of keys) {
+    if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(key)) continue;
+    if (ENV_FILE_DENYLIST.has(key)) continue;
+    if (key.startsWith("npm_") || key.startsWith("NPM_")) continue;
+    const value = env[key];
+    if (value === undefined || value === null) continue;
+    // 跳过多行值，避免 EnvironmentFile 解析失败
+    if (String(value).includes("\n") || String(value).includes("\r")) continue;
+    lines.push(`${key}=${escapeEnvFileValue(value)}`);
+  }
+  await fsp.mkdir(RUN_DIR, { recursive: true });
+  await fsp.writeFile(RUN_ENV_PATH, `${lines.join("\n")}\n`, { encoding: "utf8", mode: 0o600 });
+  await fsp.chmod(RUN_ENV_PATH, 0o600);
+}
+
+async function installSystemdUserUnit() {
+  if (!(await pathExists(SYSTEMD_UNIT_SOURCE))) {
+    throw new Error(`缺少 systemd unit 模板: ${SYSTEMD_UNIT_SOURCE}`);
+  }
+  await fsp.mkdir(SYSTEMD_USER_DIR, { recursive: true });
+  const sourceText = await fsp.readFile(SYSTEMD_UNIT_SOURCE, "utf8");
+  let needWrite = true;
+  if (await pathExists(SYSTEMD_UNIT_TARGET)) {
+    const existing = await fsp.readFile(SYSTEMD_UNIT_TARGET, "utf8");
+    needWrite = existing !== sourceText;
+  }
+  if (needWrite) {
+    await fsp.writeFile(SYSTEMD_UNIT_TARGET, sourceText, "utf8");
+    console.log(`[prod] 已安装 user unit: ${SYSTEMD_UNIT_TARGET}`);
+  }
+  runChecked("systemctl", ["--user", "daemon-reload"]);
+  // enable 保证登录/linger 后自动拉起
+  runChecked("systemctl", ["--user", "enable", SYSTEMD_UNIT_NAME]);
+}
+
+async function pointCurrentRelease(releaseDir) {
+  await fsp.mkdir(RUN_DIR, { recursive: true });
+  const tmpLink = `${RUN_CURRENT_LINK}.new-${process.pid}`;
+  try {
+    await fsp.unlink(tmpLink);
+  } catch {
+    // ignore
+  }
+  await fsp.symlink(releaseDir, tmpLink);
+  // 原子替换 current → 新 release
+  await fsp.rename(tmpLink, RUN_CURRENT_LINK);
+}
+
+function systemctlUser(args, options = {}) {
+  return runChecked("systemctl", ["--user", ...args], options);
+}
+
+function systemctlUserStatusText() {
+  try {
+    return execFileSync("systemctl", ["--user", "status", SYSTEMD_UNIT_NAME, "--no-pager"], {
+      encoding: "utf8",
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+  } catch (error) {
+    return (error.stdout || "") + (error.stderr || "") || String(error.message || error);
+  }
+}
+
+function mainPidFromSystemd() {
+  try {
+    const out = execFileSync(
+      "systemctl",
+      ["--user", "show", SYSTEMD_UNIT_NAME, "-p", "MainPID", "--value"],
+      { encoding: "utf8" },
+    ).trim();
+    const pid = Number(out);
+    return Number.isInteger(pid) && pid > 0 ? pid : null;
+  } catch {
+    return null;
+  }
+}
+
+async function stopSystemdManagedService(port) {
+  // 先停 systemd 托管（若已 enable）
+  try {
+    const active = execFileSync(
+      "systemctl",
+      ["--user", "is-active", SYSTEMD_UNIT_NAME],
+      { encoding: "utf8" },
+    ).trim();
+    if (active === "active" || active === "activating" || active === "reloading") {
+      console.log(`[prod] systemctl --user stop ${SYSTEMD_UNIT_NAME}`);
+      systemctlUser(["stop", SYSTEMD_UNIT_NAME]);
+    }
+  } catch {
+    // unit 未安装 / 未 active：忽略
+  }
+
+  // 再清掉历史 detached 裸进程 / process.json
+  await stopPreviousProcess(port);
+}
+
 async function startRelease(release) {
  const port = Number(process.env.MNOTE_PROD_PORT || DEFAULT_PORT);
  if (!Number.isInteger(port) || port <= 0 || port > 65535) {
@@ -523,11 +731,9 @@ async function startRelease(release) {
  }

  await fsp.mkdir(RUN_DIR, { recursive: true });
-  await stopPreviousProcess(port);

  const binaryPath = path.join(release.releaseDir, "mnote-web");
  const logPath = path.join(release.releaseDir, "mnote-web.log");
-  const logFd = fs.openSync(logPath, "a");
  const bind = process.env.MNOTE_WEB_BIND || `0.0.0.0:${port}`;
  const publicBind = process.env.MNOTE_WEB_PUBLIC_BIND || defaultPublicBind(port);
  const controlPlaneBackend =
@@ -535,31 +741,41 @@ async function startRelease(release) {
  if (controlPlaneBackend === "sqlite") {
    throw new Error("prod runtime 不再支持 SQLite control-plane fallback；请使用 libsql-local/turso-remote/turso-local-replica/turso-synced");
  }
+  if (!(await pathExists(binaryPath))) {
+    throw new Error(`release 二进制不存在: ${binaryPath}`);
+  }

-  const child = spawn(binaryPath, [], {
-    cwd: release.metadata?.sourceDir || release.releaseDir,
-    detached: true,
-    stdio: ["ignore", logFd, logFd],
-    env: {
-      ...process.env,
-      MNOTE_WEB_BIND: bind,
-      MNOTE_WEB_PUBLIC_BIND: publicBind,
-      MNOTE_CONTROL_PLANE_BACKEND: controlPlaneBackend,
-      ...(controlPlaneBackend === "libsql-local" || controlPlaneBackend === "turso-local" || controlPlaneBackend === "turso"
-        ? {
-            MNOTE_TURSO_LOCAL_PATH:
-              process.env.MNOTE_TURSO_LOCAL_PATH || DEFAULT_TURSO_LOCAL_PATH,
-          }
-        : {}),
-    },
-  });
+  // 1) 停旧（systemd + 裸进程）
+  await stopSystemdManagedService(port);

-  child.unref();
-  fs.closeSync(logFd);
+  // 2) 写 env / current 软链 / unit
+  const runtimeEnv = buildProdRuntimeEnv(port, bind, publicBind, controlPlaneBackend);
+  await writeSystemdEnvFile(runtimeEnv);
+  await pointCurrentRelease(release.releaseDir);
+  await installSystemdUserUnit();
+
+  // 3) 由 systemd 拉起（崩溃自动重启、linger 登录后自启）
+  console.log(`[prod] systemctl --user restart ${SYSTEMD_UNIT_NAME}`);
+  systemctlUser(["restart", SYSTEMD_UNIT_NAME]);
+
+  // 等 MainPID 出现
+  let pid = null;
+  for (let i = 0; i < 40; i += 1) {
+    pid = mainPidFromSystemd();
+    if (pid) break;
+    await sleep(250);
+  }
+  if (!pid) {
+    throw new Error(
+      `systemd 未给出 MainPID，status:\n${systemctlUserStatusText()}`,
+    );
+  }

  const state = {
-    schema: "mnote.prod_process.v1",
-    pid: child.pid,
+    schema: "mnote.prod_process.v2",
+    managedBy: "systemd-user",
+    unit: SYSTEMD_UNIT_NAME,
+    pid,
    version: release.version,
    gitHead: release.metadata?.gitHead,
    startedAt: new Date().toISOString(),
@@ -567,19 +783,25 @@ async function startRelease(release) {
    publicBind,
    releaseDir: release.releaseDir,
    binaryPath,
+    currentLink: RUN_CURRENT_LINK,
+    envPath: RUN_ENV_PATH,
    logPath,
+    journal: `journalctl --user -u ${SYSTEMD_UNIT_NAME} -f`,
  };
  await fsp.writeFile(RUN_STATE_PATH, `${JSON.stringify(state, null, 2)}\n`, "utf8");
-  await waitForReady(port, logPath, child.pid);
+
+  await waitForReady(port, logPath, pid);
  await verifyStartupAssets(port);

-  console.log(`[prod] 已启动 mnote-web release ${release.version}`);
-  console.log(`[prod] pid=${child.pid}`);
+  console.log(`[prod] 已启动 mnote-web release ${release.version}（systemd --user）`);
+  console.log(`[prod] unit=${SYSTEMD_UNIT_NAME} pid=${pid}`);
  console.log(`[prod] local URL=http://127.0.0.1:${port}`);
  for (const address of lanAddresses()) {
    console.log(`[prod] LAN URL=http://${address}:${port}`);
  }
-  console.log(`[prod] log=${logPath}`);
+  console.log(`[prod] journal: journalctl --user -u ${SYSTEMD_UNIT_NAME} -f`);
+  console.log(`[prod] log file=${logPath}（历史；新日志优先 journal）`);
+  console.log(`[prod] enable+linger: 开机/登出后仍保持运行（本机 Linger=yes）`);
 }

 async function main() {