修复 AGC 本地 BgFilter worker 启动链路 (#258)

backend 模式复用 Rust 双进程启动与 watcher

AGC 后端复用门禁增加 BgFilter worker readiness

补充启动编排与 ready 门禁回归测试

同步本地开发运维文档

---------

Co-authored-by: 段舒康 <kdletters@qq.com>
Reviewed-on: http://192.168.35.82/git/GenarrativeAI/Genarrative/pulls/258
Co-authored-by: Linghong <ink29535@proton.me>
Co-committed-by: Linghong <ink29535@proton.me>

apps/ai-game-creator-shell/scripts/start-dev-stack.mjs

@@ -79,6 +79,7 @@ function resolveBackendTargetsFromState(
 ) {
   const apiServer = state?.services?.['api-server'];
   const spacetime = state?.services?.spacetime;
+  const bgfilterWorker = state?.services?.['bgfilter-worker'];
   const isActive = (service) =>
     service && ['running', 'reused', 'starting'].includes(service.status ?? '');
   const database = typeof state?.database === 'string' ? state.database : '';
@@ -104,9 +105,14 @@ function resolveBackendTargetsFromState(
       : requireAgcBackend
         ? ''
         : 'http://127.0.0.1:3101';
+  const bgfilterWorkerUrl =
+    canReuseState && isActive(bgfilterWorker) && bgfilterWorker.url
+      ? bgfilterWorker.url
+      : '';
   return {
     apiUrl,
     spacetimeUrl,
+    bgfilterWorkerUrl,
     database,
     spacetimeDataDir,
     hasMatchingDatabase,
@@ -121,16 +127,22 @@ function readBackendTargets({ requireAgcBackend = false } = {}) {
   });
 }
 
-async function isBackendReady() {
-  const { apiUrl, spacetimeUrl, hasMatchingBackend } = readBackendTargets({
-    requireAgcBackend: true,
-  });
+async function isBackendReady({
+  state = readJson(devStackStatePath),
+  isReady = isHttpReady,
+} = {}) {
+  const { apiUrl, spacetimeUrl, bgfilterWorkerUrl, hasMatchingBackend } =
+    resolveBackendTargetsFromState(state, {
+      requireAgcBackend: true,
+    });
   return (
     hasMatchingBackend &&
     Boolean(apiUrl) &&
     Boolean(spacetimeUrl) &&
-    (await isHttpReady(`${apiUrl}/healthz`)) &&
-    (await isHttpReady(`${spacetimeUrl}/v1/ping`))
+    Boolean(bgfilterWorkerUrl) &&
+    (await isReady(`${apiUrl}/healthz`)) &&
+    (await isReady(`${spacetimeUrl}/v1/ping`)) &&
+    (await isReady(`${bgfilterWorkerUrl}/readyz`))
   );
 }
 
@@ -668,6 +680,7 @@ function isDirectModuleExecution() {
 export {
   ensureBackend,
   formatChildFailure,
+  isBackendReady,
   isDirectModuleExecution,
   isProcessGroupAlive,
   preflightExistingVite,

apps/ai-game-creator-shell/tests/start-dev-stack.test.ts

@@ -7,6 +7,7 @@ import { describe, expect, test, vi } from 'vitest';
 
 import {
   ensureBackend,
+  isBackendReady,
   isProcessGroupAlive,
   preflightExistingVite,
   readLinuxProcessGroupAlive,
@@ -21,7 +22,7 @@ import {
 const expectedDatabase = 'genarrative-game-creator-dev';
 const expectedDataDir = resolve('server-rs/.spacetimedb/ai-game-creator/data');
 
-function backendState(spacetimeDataDir?: string) {
+function backendState(spacetimeDataDir?: string, includeBgfilterWorker = true) {
   return {
     schemaVersion: spacetimeDataDir ? 2 : 1,
     database: expectedDatabase,
@@ -35,6 +36,14 @@ function backendState(spacetimeDataDir?: string) {
         status: 'running',
         url: 'http://127.0.0.1:3101',
       },
+      ...(includeBgfilterWorker
+        ? {
+            'bgfilter-worker': {
+              status: 'running',
+              url: 'http://127.0.0.1:8083',
+            },
+          }
+        : {}),
     },
   };
 }
@@ -87,6 +96,36 @@ describe('AI 游戏创作配套后端复用门禁', () => {
     expect(matching.hasMatchingBackend).toBe(true);
     expect(matching.apiUrl).toBe('http://127.0.0.1:8082');
     expect(matching.spacetimeUrl).toBe('http://127.0.0.1:3101');
+    expect(matching.bgfilterWorkerUrl).toBe('http://127.0.0.1:8083');
+  });
+
+  test('worker 缺失或未 ready 时不允许复用后端', async () => {
+    const isReady = vi.fn(async (_url: string) => true);
+
+    await expect(
+      isBackendReady({
+        state: backendState(expectedDataDir, false),
+        isReady,
+      }),
+    ).resolves.toBe(false);
+    expect(isReady).not.toHaveBeenCalled();
+
+    isReady.mockImplementation(async (url) => url.endsWith('/readyz'));
+    await expect(
+      isBackendReady({
+        state: backendState(expectedDataDir),
+        isReady,
+      }),
+    ).resolves.toBe(false);
+    expect(isReady).toHaveBeenCalledWith('http://127.0.0.1:8082/healthz');
+
+    isReady.mockImplementation(async () => true);
+    await expect(
+      isBackendReady({
+        state: backendState(expectedDataDir),
+        isReady,
+      }),
+    ).resolves.toBe(true);
   });
 });
 

docs/【开发运维】本地开发验证与生产运维-2026-05-15.md

@@ -58,7 +58,7 @@ Linux 本机多用户并发开发时，`npm run dev`、`npm run dev:*` 单模块
 
 后端日志默认写入 `logs/api-server/`，独立 BgFilter worker 日志默认写入 `logs/bgfilter-worker/`。后端 API smoke 使用 `npm run dev:api-server`，先检查 BgFilter worker `/readyz`，再检查 API `/healthz`；需要确认 API 实例可接生产流量时检查 API `/readyz`。不要使用旧 `api-server:maincloud` 或任何 `GENARRATIVE_SPACETIME_MAINCLOUD_*` 口径。
 
-AI 游戏创作客户端使用 `npm run agc`。该入口由 `apps/ai-game-creator-shell/scripts/start-tauri-dev.mjs` 解析 AGC Vite 实际端口：Linux 默认取当前用户端口段的 `start + 5`，占用时只在本用户段内漂移；Windows / macOS 保留 `3080` 为兼容首选并允许统一漂移。最终端口通过 `GENARRATIVE_AGC_VITE_PORT` 传给 `beforeDevCommand` 和配套后端端口解析器，通过 Tauri CLI 动态 `build.devUrl` 配置传给 WebView，并通过 Vite CLI `--port` 启动严格监听；Vite 继续使用 `strictPort`，任何一层都不得自行改到另一个端口。启动器在创建原生窗口前预检最终地址；若竞态中该地址被 AGC Vite、无响应监听器或其它服务占用，一律失败关闭，不复用、也不擅自终止无法证明归属的进程。
+AI 游戏创作客户端使用 `npm run agc`。该入口由 `apps/ai-game-creator-shell/scripts/start-tauri-dev.mjs` 解析 AGC Vite 实际端口：Linux 默认取当前用户端口段的 `start + 5`，占用时只在本用户段内漂移；Windows / macOS 保留 `3080` 为兼容首选并允许统一漂移。最终端口通过 `GENARRATIVE_AGC_VITE_PORT` 传给 `beforeDevCommand` 和配套后端端口解析器，通过 Tauri CLI 动态 `build.devUrl` 配置传给 WebView，并通过 Vite CLI `--port` 启动严格监听；Vite 继续使用 `strictPort`，任何一层都不得自行改到另一个端口。AGC 配套后端的 `backend` 模式启动 SpacetimeDB、独立 `bgfilter-worker` 和 `api-server`，并在复用现有后端前同时检查三者状态及 `/v1/ping`、`/readyz`、`/healthz`；worker 缺失时不得把不完整的 API/数据库组合误判为 ready。启动器在创建原生窗口前预检最终地址；若竞态中该地址被 AGC Vite、无响应监听器或其它服务占用，一律失败关闭，不复用、也不擅自终止无法证明归属的进程。
 
 Tauri `beforeDevCommand` 默认与客户端构建并行，不能把上述检查只放在 `beforeDevCommand` 内：选定地址上若已有旧 Vite，Tauri 可能先创建加载旧前端的窗口，随后配套后端才因代理不匹配退出。外层启动器会把 Tauri CLI 放入受控进程树；CLI 正常退出、启动失败或收到终止信号后，POSIX 先向保留的 PGID 发送 `SIGTERM`、有界等待后升级 `SIGKILL`，Windows 使用 `taskkill /PID <pid> /T /F`。Linux 容器中的孤儿后代退出后可能暂时保留为 zombie，`kill(-PGID, 0)` 仍会返回成功；启动器必须结合 `/proc/<pid>/stat` 判断同组是否还存在非 zombie 成员，不能把等待 PID 1 回收误报为清理失败。配套后端和 Vite 仍由 `start-dev-stack.mjs` 各自持有，退出时同样有界收束，避免只剩客户端、Runner、Cargo 或旧订阅进程。排障时同时核对控制台输出的 AGC Vite 实际地址及其 marker、`.app/dev-stack.json` 的实际 API URL 和进程 cwd；不要把“终端已返回”当成客户端及其 Runner 已退出的证据。
 

scripts/dev.mjs

@@ -1619,9 +1619,8 @@ class DevRunner {
 
     if (command === 'backend') {
       await this.startSpacetimeForFullStack();
-      await this.services.get('api-server').start();
-      await this.waitForApiServer();
-      this.startWatchers(['spacetime', 'api-server']);
+      await this.startRustServicePair();
+      this.startWatchers(['spacetime', 'api-server', 'bgfilter-worker']);
       return;
     }
 

scripts/dev.test.ts

@@ -500,6 +500,30 @@ describe('dev scheduler Rust service orchestration', () => {
     ]);
   });
 
+  test('backend 模式复用 Rust 双进程并纳入三服务 watcher', async () => {
+    const { explicitOptions, options } = parseArgs(['backend'], {});
+    const runner = new DevRunner(options, {}, explicitOptions);
+    const startSpacetime = vi
+      .spyOn(runner, 'startSpacetimeForFullStack')
+      .mockResolvedValue(undefined);
+    const startPair = vi
+      .spyOn(runner, 'startRustServicePair')
+      .mockResolvedValue(undefined);
+    const startWatchers = vi
+      .spyOn(runner, 'startWatchers')
+      .mockImplementation(() => {});
+
+    await runner.startCommand('backend');
+
+    expect(startSpacetime).toHaveBeenCalledOnce();
+    expect(startPair).toHaveBeenCalledOnce();
+    expect(startWatchers).toHaveBeenCalledWith([
+      'spacetime',
+      'api-server',
+      'bgfilter-worker',
+    ]);
+  });
+
   test('完整栈只为两个 Rust 角色创建一套组合 watcher', () => {
     const { explicitOptions, options } = parseArgs(['--watch'], {});
     const runner = new DevRunner(options, {}, explicitOptions);