优化启动流程

2026-04-13 19:01:48 +08:00
parent 7008c85e58
commit dc8a014601
18 changed files with 837 additions and 17 deletions
@@ -10,15 +10,15 @@ import { ensureAdminUsersBootstrapped } from './services/admin/admin-auth-servic
 import { ensureFulfillmentCatalogBootstrapped } from './services/bootstrap/fulfillment-bootstrap-service.js'
 import { closeLocalOcrWorker, warmupLocalOcrWorker } from './services/session/ocr.js'
 import { closeAllTencentBrowserSessions, warmupTencentBrowser } from './services/session/session.js'
+import { buildSuccessPayload } from './utils/http.js'
 import { logError, logInfo, logWarn } from './utils/logger.js'
 import tencentRouter from './routes/tencent.js'

+const CORE_BOOT_RETRY_DELAY_MS = 5_000
+
 const app = express()
 const port = Number(runtimeConfig.server.port || 3000)
-
-await runDatabaseMigrations()
-await ensureFulfillmentCatalogBootstrapped()
-await ensureAdminUsersBootstrapped()
+const startupState = createStartupState()

 app.use((req, res, next) => {
  res.setHeader('Access-Control-Allow-Origin', '*')
@@ -37,10 +37,49 @@ app.use(express.json({ limit: '2mb' }))
 app.use(express.urlencoded({ extended: true }))

 app.get('/health', (_req, res) => {
-  res.json({
-    code: 0,
-    msg: 'ok',
+  res.json(buildSuccessPayload(buildHealthPayload(), startupState.core.ready ? 'ready' : 'starting'))
+})
+
+app.get('/health/live', (_req, res) => {
+  res.json(buildSuccessPayload({
+    status: shutdownStarted ? 'shutting_down' : 'alive',
+    pid: process.pid,
+  }))
+})
+
+app.get('/health/ready', (_req, res) => {
+  if (startupState.core.ready) {
+    res.json(buildSuccessPayload(buildHealthPayload(), 'ready'))
+    return
+  }
+
+  res.status(503).json({
+    code: 1,
+    msg: startupState.core.lastError || '服务启动中，请稍后重试',
+    errorCode: 'service_not_ready',
    time: Math.floor(Date.now() / 1000),
+    data: buildHealthPayload(),
+  })
+})
+
+app.use((req, res, next) => {
+  if (startupState.core.ready) {
+    next()
+    return
+  }
+
+  res.status(503).json({
+    code: 1,
+    msg: startupState.core.lastError || '服务启动中，请稍后重试',
+    errorCode: 'service_not_ready',
+    time: Math.floor(Date.now() / 1000),
+    data: {
+      startup: {
+        phase: startupState.phase,
+        attemptCount: startupState.core.attemptCount,
+        lastAttemptAt: startupState.core.lastAttemptAt,
+      },
+    },
  })
 })

@@ -51,23 +90,85 @@ app.use('/api/v1/admin', adminRouter)

 const server = app.listen(port, () => {
  logInfo('[startup]', `order-site-backend listening on http://127.0.0.1:${port}`)
-  void bootstrapBrowser()
-  void bootstrapOcr()
+  void bootstrapCoreServices()
 })

+server.on('error', (error) => {
+  logError('[startup]', 'HTTP server failed', error)
+})
+
+async function bootstrapCoreServices() {
+  if (startupState.core.running || shutdownStarted) {
+    return
+  }
+
+  startupState.core.running = true
+
+  while (!shutdownStarted && !startupState.core.ready) {
+    startupState.phase = startupState.core.attemptCount === 0 ? 'starting' : 'retrying'
+    startupState.core.attemptCount += 1
+    startupState.core.lastAttemptAt = new Date().toISOString()
+
+    try {
+      logInfo('[startup]', '开始执行核心启动步骤', {
+        attempt: startupState.core.attemptCount,
+      })
+
+      await runDatabaseMigrations()
+      await ensureFulfillmentCatalogBootstrapped()
+      await ensureAdminUsersBootstrapped()
+
+      startupState.core.ready = true
+      startupState.core.lastError = ''
+      startupState.phase = 'ready'
+      startupState.core.readyAt = new Date().toISOString()
+
+      logInfo('[startup]', '核心启动步骤完成，服务已就绪', {
+        attempt: startupState.core.attemptCount,
+      })
+
+      void bootstrapBrowser()
+      void bootstrapOcr()
+      break
+    } catch (error) {
+      const message = formatStartupError(error)
+      startupState.phase = 'retrying'
+      startupState.core.lastError = message
+      logError('[startup]', '核心启动步骤失败，将自动重试', {
+        attempt: startupState.core.attemptCount,
+        retryDelayMs: CORE_BOOT_RETRY_DELAY_MS,
+        error,
+      })
+
+      await sleep(CORE_BOOT_RETRY_DELAY_MS)
+    }
+  }
+
+  startupState.core.running = false
+}
+
 async function bootstrapBrowser() {
+  startupState.browser.status = 'starting'
+  startupState.browser.lastAttemptAt = new Date().toISOString()
+
  try {
    const result = await warmupTencentBrowser()

    if (!result.warmed) {
+      startupState.browser.status = 'skipped'
+      startupState.browser.message = 'browser prewarm skipped'
      logInfo('[startup]', 'browser prewarm skipped')
      return
    }

+    startupState.browser.status = 'ready'
+    startupState.browser.message = 'browser prewarm ready'
    logInfo('[startup]', 'browser prewarm ready')
  } catch (error) {
    if (isMissingPlaywrightBrowserError(error)) {
      const installCommand = resolveBrowserInstallCommand()
+      startupState.browser.status = 'degraded'
+      startupState.browser.message = formatErrorMessage(error)
      logWarn('[startup]', `browser prewarm skipped: ${formatErrorMessage(error)}`)
      logWarn('[startup]', `请先安装浏览器依赖: ${installCommand}`)

@@ -78,19 +179,88 @@ async function bootstrapBrowser() {
      return
    }

+    startupState.browser.status = 'degraded'
+    startupState.browser.message = formatStartupError(error)
    logError('[startup]', 'browser prewarm failed', error)
  }
 }

 async function bootstrapOcr() {
+  startupState.ocr.status = 'starting'
+  startupState.ocr.lastAttemptAt = new Date().toISOString()
+
  try {
    await warmupLocalOcrWorker()
+    startupState.ocr.status = 'ready'
+    startupState.ocr.message = 'local OCR ready'
    logInfo('[startup]', 'local OCR ready')
  } catch (error) {
+    startupState.ocr.status = 'degraded'
+    startupState.ocr.message = formatStartupError(error)
    logWarn('[startup]', `local OCR skipped: ${formatStartupError(error)}`)
  }
 }

+function buildHealthPayload() {
+  return {
+    phase: startupState.phase,
+    startedAt: startupState.startedAt,
+    core: {
+      ready: startupState.core.ready,
+      running: startupState.core.running,
+      attemptCount: startupState.core.attemptCount,
+      readyAt: startupState.core.readyAt,
+      lastAttemptAt: startupState.core.lastAttemptAt,
+      lastError: startupState.core.lastError,
+    },
+    browser: {
+      status: startupState.browser.status,
+      message: startupState.browser.message,
+      lastAttemptAt: startupState.browser.lastAttemptAt,
+    },
+    ocr: {
+      status: startupState.ocr.status,
+      message: startupState.ocr.message,
+      lastAttemptAt: startupState.ocr.lastAttemptAt,
+    },
+    process: {
+      pid: process.pid,
+      shutdownStarted,
+      lastUnhandledRejection: startupState.process.lastUnhandledRejection,
+      lastUncaughtException: startupState.process.lastUncaughtException,
+    },
+  }
+}
+
+function createStartupState() {
+  return {
+    phase: 'starting',
+    startedAt: new Date().toISOString(),
+    core: {
+      ready: false,
+      running: false,
+      attemptCount: 0,
+      readyAt: '',
+      lastAttemptAt: '',
+      lastError: '',
+    },
+    browser: {
+      status: 'pending',
+      message: '',
+      lastAttemptAt: '',
+    },
+    ocr: {
+      status: 'pending',
+      message: '',
+      lastAttemptAt: '',
+    },
+    process: {
+      lastUnhandledRejection: null,
+      lastUncaughtException: null,
+    },
+  }
+}
+
 function formatStartupError(error) {
  if (error instanceof Error && error.message) {
    return error.message.split('\n')[0].trim()
@@ -126,6 +296,12 @@ function resolveBrowserInstallCommand() {
  return 'npm run browser:install'
 }

+function sleep(ms) {
+  return new Promise((resolve) => {
+    setTimeout(resolve, ms)
+  })
+}
+
 let shutdownStarted = false

 async function shutdown(signal) {
@@ -134,6 +310,7 @@ async function shutdown(signal) {
  }

  shutdownStarted = true
+  startupState.phase = 'shutting_down'
  logInfo('[shutdown]', `received ${signal}, closing browser sessions and HTTP server`)

  try {
@@ -171,3 +348,19 @@ process.on('SIGINT', () => {
 process.on('SIGTERM', () => {
  void shutdown('SIGTERM')
 })
+
+process.on('unhandledRejection', (reason) => {
+  startupState.process.lastUnhandledRejection = {
+    time: new Date().toISOString(),
+    message: formatStartupError(reason),
+  }
+  logError('[process]', 'unhandled promise rejection', reason)
+})
+
+process.on('uncaughtException', (error) => {
+  startupState.process.lastUncaughtException = {
+    time: new Date().toISOString(),
+    message: formatStartupError(error),
+  }
+  logError('[process]', 'uncaught exception captured', error)
+})