mslxdff 0.1.187 → 0.1.188

package/src/cli/commands/system.js

@@ -3,7 +3,7 @@ import { join, dirname } from "node:path";
 import { fileURLToPath } from "node:url";
 import { defaultStateFile } from "../../state.js";
 import { loadToken, refreshToken } from "../../state.js";
-import { stopDaemon, pidFile, logFile } from "../../daemon.js";
+import { stopDaemon, pidFile, logFile, isPidAlive } from "../../daemon.js";
 import { logDir, eventsFile, callsFile, errorsFile, recentEvents, timelineFile, recentTimeline } from "../../logs.js";
 import { fmtEvent } from "../format.js";
 import { fmtShanghaiYMDHMS, fmtShanghaiHMS } from "../../time.js";
@@ -58,10 +58,21 @@ export async function handleUpdate(args, VERSION) {
   const daemon = readPid();
   if (daemon) {
     console.log("restarting daemon on the new version…");
-    stopDaemon({ reason: "restart" });
-    startDaemon([]);
-    await waitForHealth(resolvePort(), 4000);
-    console.log(`restarted (pid ${readPid()})`);
+    const r = stopDaemon({ reason: "restart" });
+    // 同 BUG-29：停不掉就绝不能报 restarted —— 旧版还占着端口，新进程 EADDRINUSE 秒退，
+    // 而 waitForHealth 打的是旧 daemon 的 /health，照样 200，读 pid 读到的也是旧值。
+    if (!r.stopped && r.pid) {
+      console.error(`⚠ 旧 daemon (pid ${r.pid}) 没能停掉：${r.reason}`);
+      console.error(`  新版本已装好，但旧版仍在运行（未生效）。请用管理员终端停掉它后重新 -restart。`);
+      process.exit(1);
+    }
+    const spawned = startDaemon([]);
+    const health = await waitForHealth(resolvePort(), 4000, { expectPid: spawned });
+    if (!isPidAlive(spawned)) {
+      console.error(`⚠ 新 daemon (pid ${spawned}) 启动后立即退出 —— 端口大概仍被旧进程占着（EADDRINUSE）。旧版仍在运行，升级未生效。`);
+      process.exit(1);
+    }
+    console.log(`restarted (pid ${spawned})${health.ok ? "" : " — health check pending"}`);
   }
   process.exit(0);
 }
@@ -82,9 +93,14 @@ export async function handleShowToken(args) {
 
 export async function handleUninstall(args) {
   if (!(args.includes("-uninstall") || args.includes("--uninstall"))) return false;
-  const { stopped, pid } = stopDaemon({ reason: "uninstall" });
+  const { stopped, pid, reason } = stopDaemon({ reason: "uninstall" });
   if (stopped) console.log(`mslxdff daemon stopped (pid ${pid})`);
-  else console.log("mslxdff daemon not running");
+  else if (pid) {
+    // 杀被拒 ≠ 没在跑：它还活着并持有 state/logs，继续删文件会让在跑的 daemon 半瘫（写不进也不重来）。
+    console.error(`mslxdff daemon still running (pid ${pid}) — uninstall aborted: ${reason}`);
+    console.error("先停掉它（管理员终端执行 mslxdff -stop，或 taskkill /PID " + pid + " /F），再重新 -uninstall");
+    process.exit(1);
+  } else console.log("mslxdff daemon not running");
   const stateFile = defaultStateFile();
   const dir = dirname(stateFile);
   const removed = [];

package/src/cli/join-wizard.js

@@ -52,8 +52,11 @@ export async function ensureServiceRunning() {
   const pid = readPid();
   if (pid && isPidAlive(pid)) return { pid, started: false };
   const newPid = startDaemon([]);
-  await waitForHealth(effectivePort([]), SERVICE_HEALTH_TIMEOUT_MS);
-  return { pid: newPid, started: true };
+  // 只认刚起的 pid：旧 daemon 若仍占端口，它的 /health 照样 200（同 BUG-29/30）。
+  const health = await waitForHealth(effectivePort([]), SERVICE_HEALTH_TIMEOUT_MS, { expectPid: newPid });
+  // 起完立即退出 = 端口被占（EADDRINUSE）。向导不能报「服务已启动」，那会把后续入组失败归错因。
+  if (!isPidAlive(newPid)) return { pid: null, started: false, failed: true, reason: "daemon exited right after start (port in use?)" };
+  return { pid: newPid, started: true, healthy: health.ok };
 }
 
 /**
@@ -148,8 +151,9 @@ export async function runJoinWizard(opts = {}) {
 
   out("→ 正在启动后台服务 ...");
   const service = await ensureService().catch((err) => ({ error: errMsg(err) }));
-  if (service?.error) {
-    out(`✗ 后台服务启动失败：${service.error}`);
+  if (service?.error || service?.failed) {
+    // failed 是「进程起来又立刻死了」（端口被占用）：不能落到下面「已在运行（pid null）」分支谎报成功。
+    out(`✗ 后台服务启动失败：${service.error || service.reason}`);
     out("  手动启动：mslxdff -d      查看日志：mslxdff -log 50");
   } else if (service?.started) {
     out(`✓ 后台服务已启动（pid ${service.pid}）`);

package/src/cli/policy.js

@@ -36,17 +36,29 @@ export function effectiveHost() {
   return "0.0.0.0";
 }
 
-export async function waitForHealth(port, timeoutMs) {
+/**
+ * 等 /health 可用。给了 expectPid 就**只认那个 pid**：旧 daemon 仍占着端口时它也会回 200，
+ * 不校验身份的话「新进程 EADDRINUSE 秒退、旧版还在服务」会被当成启动成功（BUG-29）。
+ * 返回 { ok, pid, version }：ok=true 表示确认到位（无 expectPid 时即 res.ok）。
+ */
+export async function waitForHealth(port, timeoutMs, { expectPid = null } = {}) {
   const start = Date.now();
   while (Date.now() - start < timeoutMs) {
     try {
       const res = await compatFetch(`http://127.0.0.1:${port}/health`);
-      if (res.ok) return;
+      if (res.ok) {
+        if (expectPid == null) return { ok: true, pid: null, version: null };
+        let body = null;
+        try { body = await res.json(); } catch { /* 老版本 daemon 不认 JSON 之外的东西 */ }
+        // 老版本 /health 只回 {status:"ok"}，没有 pid ⇒ 无法证明是新进程，继续等（旧版本来就该被换掉）
+        if (body && Number(body.pid) === Number(expectPid)) return { ok: true, pid: body.pid, version: body.version ?? null };
+      }
     } catch {
       // not up yet
     }
     await new Promise((r) => setTimeout(r, 100));
   }
+  return { ok: false, pid: null, version: null };
 }
 
 // Only upgrade, never downgrade. Keeps a newer daemon running.