fix: wait for slow ARMv6 scheduler workers to become ready

back/app.ts

@@ -41,11 +41,18 @@ class Application {
   }
 
   private startMasterProcess() {
+    const readyTimeout =
+      process.arch === 'arm' &&
+      Number(
+        (process.config.variables as { arm_version?: string | number }).arm_version,
+      ) === 6
+        ? 120000
+        : 30000;
     // Fork gRPC worker first and wait for it to be ready
     const grpcWorker = this.forkWorker('grpc');
     
     // Wait for gRPC worker to signal it's ready before starting HTTP worker
-    this.waitForWorkerReady(grpcWorker, 30000)
+    this.waitForWorkerReady(grpcWorker, readyTimeout)
       .then(() => {
         Logger.info('[boot] gRPC worker is ready, starting HTTP worker');
         this.httpWorker = this.forkWorker('http');
@@ -71,7 +78,7 @@ class Application {
               Logger.warn('Unable to notify HTTP worker of scheduler exit');
             }
             const newGrpcWorker = this.forkWorker('grpc');
-            this.waitForWorkerReady(newGrpcWorker, 30000)
+            this.waitForWorkerReady(newGrpcWorker, readyTimeout)
               .then(() => {
                 Logger.info('gRPC worker restarted and ready');
                 // Re-register cron jobs by notifying the HTTP worker

docs/development/platform-upgrade-bundles.md

@@ -45,7 +45,7 @@ ARMv6 glibc 小设备（例如 Raspberry Pi 1/Zero 的 Raspbian/Raspberry Pi OS
 - glibc、musl 分包，并记录构建系统 libc 下限。系统库仍以目标设备加载测试作为最后检查。
 - ARMv6 glibc 使用签名校验的 Raspbian bullseye（glibc 2.31）构建环境，以固定 SHA256 的 ARMv6 Node 20.18.1 头文件编译 SQLite，强制 `-march=armv6 -mfpu=vfp -mfloat-abi=hard`。CI 检查 ELF 的 ARMv6 和硬浮点标记，并在 QEMU ARM1176 下完成归档复验、SQLite 读写和空目录离线安装；验证辅助镜像按不可变摘要记录，不进入用户安装包。Debian ARMv7 镜像仅提供同次构建的应用文件，不复用其 Node、libc 或 SQLite 二进制。
 - ARMv6 与 ARMv7 使用不同包名。ARMv6 在目标镜像中从源码编译 SQLite，禁止复用 generic `arm` 的 ARMv7 预编译件；构建需通过加载和内存库读写，失败则不发布渠道。
-- ARMv6 的归档构建、在线更新工作进程及独立离线安装器使用 `--jitless`，避开 ARM1176 上 V8 的优化编译崩溃；保留现有 Node 二进制，正常面板进程的启动参数不变。ARMv6 升级后的健康检查最多等待 5 分钟，覆盖慢设备按顺序启动数据库、调度与 HTTP 工作进程的耗时；其他平台仍为 60 秒，超时或版本校验失败仍执行回退。依赖加载和 SQLite 读写仍由现有 Node 的正常模式独立进程验证，使 Node 20.18.1 的内置 HTTP/WebAssembly 组件保持可用。
+- ARMv6 的归档构建、在线更新工作进程及独立离线安装器使用 `--jitless`，避开 ARM1176 上 V8 的优化编译崩溃；保留现有 Node 二进制，正常面板进程的启动参数不变。ARMv6 的 gRPC 工作进程首次启动及重建最多等待 2 分钟；升级后的整体健康检查最多等待 5 分钟，覆盖慢设备按顺序启动数据库、调度与 HTTP 工作进程的耗时；其他平台仍为 60 秒，超时或版本校验失败仍执行回退。依赖加载和 SQLite 读写仍由现有 Node 的正常模式独立进程验证，使 Node 20.18.1 的内置 HTTP/WebAssembly 组件保持可用。
 - tar 中允许 pnpm 的包内符号链接；拒绝绝对路径、越界路径、包外链接、硬链接、设备条目和经符号链接写入的条目。重启前再次校验依赖，暂存失败保留上一次可用的就绪包。
 
 自托管镜像可设置 `QL_UPGRADE_INDEX_URL` 指向兼容索引；要求 HTTPS，测试时仅放行 loopback HTTP。索引和所有平台归档应部署到同一可访问源，避免依赖额外的 raw 域名。

test/back/armv6-worker-readiness.test.cjs

@@ -0,0 +1,101 @@
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const { EventEmitter } = require('node:events');
+const ts = require('typescript');
+
+const source = ts.createSourceFile(
+  'app.ts',
+  fs.readFileSync('back/app.ts', 'utf8'),
+  ts.ScriptTarget.Latest,
+  true,
+);
+const application = source.statements.find(
+  (node) => ts.isClassDeclaration(node) && node.name.text === 'Application',
+);
+const method = application.members.find(
+  (node) => node.name?.getText(source) === 'startMasterProcess',
+);
+const compiled = ts.transpileModule(
+  `class Application { ${method.getText(
+    source,
+  )} }\nmodule.exports = Application;`,
+  {
+    compilerOptions: {
+      module: ts.ModuleKind.CommonJS,
+      target: ts.ScriptTarget.ES2020,
+    },
+  },
+).outputText;
+
+for (const [arch, armVersion, timeout] of [
+  ['arm', 6, 120000],
+  ['arm', 7, 30000],
+  ['arm64', undefined, 30000],
+]) {
+  test(`${arch} v${
+    armVersion ?? ''
+  } waits for scheduler readiness before HTTP and job recovery`, async () => {
+    const cluster = new EventEmitter();
+    const module = { exports: {} };
+    const logger = { info() {}, warn() {}, error() {} };
+    new Function(
+      'module',
+      'cluster',
+      'Logger',
+      'errStack',
+      'process',
+      compiled,
+    )(module, cluster, logger, String, {
+      arch,
+      config: { variables: { arm_version: armVersion } },
+      exit() {
+        assert.fail('unexpected application exit');
+      },
+    });
+    const app = new module.exports();
+    app.workerMetadataMap = new Map();
+    const forks = [],
+      waits = [],
+      notifications = [];
+    let ready;
+    app.forkWorker = (serviceType) => {
+      const worker = {
+        id: forks.length + 1,
+        process: { pid: 100 + forks.length },
+        send: (message) => notifications.push(message),
+      };
+      forks.push({ serviceType, worker });
+      app.workerMetadataMap.set(worker.id, { serviceType });
+      return worker;
+    };
+    app.waitForWorkerReady = (worker, milliseconds) => {
+      waits.push({ worker, milliseconds });
+      return new Promise((resolve) => {
+        ready = resolve;
+      });
+    };
+    app.setupMasterShutdown = () => {};
+    app.startMasterProcess();
+    assert.deepEqual(
+      forks.map((fork) => fork.serviceType),
+      ['grpc'],
+    );
+    assert.equal(waits[0].milliseconds, timeout);
+    ready();
+    await Promise.resolve();
+    assert.deepEqual(
+      forks.map((fork) => fork.serviceType),
+      ['grpc', 'http'],
+    );
+    cluster.emit('exit', forks[0].worker, 1);
+    assert.equal(waits[1].milliseconds, timeout);
+    assert.deepEqual(notifications, ['scheduler-unavailable']);
+    ready();
+    await Promise.resolve();
+    assert.deepEqual(notifications, [
+      'scheduler-unavailable',
+      'reregister-crons',
+    ]);
+  });
+}