agentworkshop 0.8.6 → 0.8.8

package/server/services/workshop/daq/infra.ts

@@ -1,229 +1,250 @@
-/**
- * DAQ 基础设施编排 —— 配置驱动的真实 MQTT/Timescale 生命周期。
- *
- * 启动策略(config.yml daq.startInfrastructure):
- *   auto   = 探测配置地址;端口不通且宿主有 Docker → docker compose 自动拉起,
- *            拉起后等待就绪;无 Docker → 直连失败即降级;
- *   always = 仅经 Docker 拉起(不探测直连);
- *   never  = 只按配置直连。
- *
- * 降级语义:任一后端不可达 → 该后端回退 mock 对应物(进程内队列 / SQLite 时序仿真),
- * 控制器停采(在线数采停用),meta 携带 warning,前端横幅 + 一键重连;
- * 后台每 30s 自动重试,恢复后自动重启采集并切回真实后端。
- */
-import { createLogger } from '../logger'
-import { execFile } from 'node:child_process'
-import { createConnection } from 'node:net'
-import { promisify } from 'node:util'
-import { join } from 'node:path'
-
-const log = createLogger('daq.infra')
-
-const execFileAsync = promisify(execFile)
-
-export interface DaqInfraConfig {
-  startInfrastructure: 'auto' | 'always' | 'never'
-  mqtt: { host: string, port: number, username?: string, password?: string, secure?: boolean, caFile?: string }
-  timescale: { host: string, port: number, user: string, password: string, database: string }
-  /** 对象存储(v2:图像帧;不可达 → 本地磁盘降级,不阻断采集) */
-  objectstore?: { host: string, port: number, accessKey: string, secretKey: string, bucket: string }
-}
-
-export interface DaqInfraStatus {
-  /** 真实 MQTT 是否在线(离线 → 进程内队列降级) */
-  mqttOnline: boolean
-  /** Timescale 是否在线(离线 → SQLite 时序仿真降级) */
-  tsdbOnline: boolean
-  /** 对象存储是否在线(离线 → 本地磁盘降级;不计入 degraded,采集不中断) */
-  objectStoreOnline: boolean
-  /** 任一降级 → 在线采集停用,前端横幅警告 */
-  degraded: boolean
-  warning: string
-  /** 基础设施拉起方式:docker / direct / none */
-  startedBy: 'docker' | 'direct' | 'none'
-  lastCheckAt: string
-  lastError?: string
-}
-
-const g = globalThis as typeof globalThis & {
-  __daqInfra?: { cfg: DaqInfraConfig, status: DaqInfraStatus, retryTimer?: NodeJS.Timeout, projectRoot?: string }
-}
-
-function state() {
-  return g.__daqInfra ??= {
-    cfg: { startInfrastructure: 'auto', mqtt: { host: '127.0.0.1', port: 1883 }, timescale: { host: '127.0.0.1', port: 5432, user: 'postgres', password: 'awshop', database: 'awshop' } },
-    status: { mqttOnline: false, tsdbOnline: false, objectStoreOnline: false, degraded: true, warning: '', startedBy: 'none', lastCheckAt: new Date().toISOString() },
-  }
-}
-
-/** TCP 端口探测(2s 超时;Docker 端口映射即代表服务可连) */
-function probePort(host: string, port: number, timeoutMs = 2000): Promise<boolean> {
-  return new Promise((resolve) => {
-    const sock = createConnection({ host, port })
-    const done = (ok: boolean): void => {
-      sock.destroy()
-      resolve(ok)
-    }
-    sock.setTimeout(timeoutMs, () => done(false))
-    sock.once('connect', () => done(true))
-    sock.once('error', () => done(false))
-  })
-}
-
-async function dockerAvailable(): Promise<boolean> {
-  try {
-    await execFileAsync('docker', ['--version'], { timeout: 5000 })
-    return true
-  }
-  catch {
-    return false
-  }
-}
-
-async function dockerComposeUp(projectRoot: string): Promise<{ ok: boolean, error?: string }> {
-  try {
-    await execFileAsync('docker', ['compose', '-f', join(projectRoot, 'docker-compose.yml'), 'up', '-d', 'daq-mosquitto', 'daq-timescale', 'daq-minio'], { timeout: 180_000, cwd: projectRoot })
-    return { ok: true }
-  }
-  catch (err) {
-    return { ok: false, error: err instanceof Error ? err.message : String(err) }
-  }
-}
-
-/** 等待三个端口就绪(拉起后服务初始化需要数秒;最长 waitMs) */
-async function waitPorts(cfg: DaqInfraConfig, waitMs: number): Promise<{ mqtt: boolean, tsdb: boolean, os: boolean }> {
-  const deadline = Date.now() + waitMs
-  let mqtt = false
-  let tsdb = false
-  let os = false
-  while (Date.now() < deadline && !(mqtt && tsdb && os)) {
-    mqtt = mqtt || await probePort(cfg.mqtt.host, cfg.mqtt.port, 1200)
-    tsdb = tsdb || await probePort(cfg.timescale.host, cfg.timescale.port, 1200)
-    os = os || await probePort(cfg.objectstore?.host ?? '127.0.0.1', cfg.objectstore?.port ?? 9000, 1200)
-    if (!(mqtt && tsdb && os)) await new Promise(r => setTimeout(r, 1500))
-  }
-  return { mqtt, tsdb, os }
-}
-
-export function daqInfraStatus(): DaqInfraStatus {
-  return { ...state().status }
-}
-
-/**
- * 后端装配失败(S10):端口探测通过但 TSDB/队列/对象存储 init 失败时,
- * 必须把状态显式置为 degraded —— 否则 30s 自动重连永不触发,采样会写进占位后端。
- */
-export function markDaqInfraFailure(err: unknown): void {
-  const s = state().status
-  s.degraded = true
-  s.lastCheckAt = new Date().toISOString()
-  s.lastError = err instanceof Error ? err.message : String(err)
-  s.warning = `数采后端初始化失败,已降级并停采:${s.lastError}`
-  log.error('[daq-infra]', s.warning)
-}
-
-/** 运行配置快照(启动插件装配时写入;重连端点复用) */
-export function daqInfraConfig(): DaqInfraConfig {
-  return state().cfg
-}
-
-/** 配置 → 连接 URL(env 显式覆盖优先,便于外部托管 broker/db/对象存储;S1:支持凭据/TLS) */
-export function daqUrls(cfg = state().cfg): { mqttUrl: string | null, tsdbUrl: string | null, objectStoreUrl: string | null } {
-  // S1:凭据/TLS 优先 env(DAQ_MQTT_URL 整串覆盖),否则由 config.mqtt 拼(含 user:pass@ 与 mqtts://)
-  let mqttUrl: string
-  if (process.env.DAQ_MQTT_URL) {
-    mqttUrl = process.env.DAQ_MQTT_URL
-  }
-  else {
-    const scheme = cfg.mqtt.secure ? 'mqtts' : 'mqtt'
-    const cred = cfg.mqtt.username
-      ? `${encodeURIComponent(cfg.mqtt.username)}:${encodeURIComponent(cfg.mqtt.password ?? '')}@`
-      : ''
-    mqttUrl = `${scheme}://${cred}${cfg.mqtt.host}:${cfg.mqtt.port}`
-  }
-  const { user, password, database, host, port } = cfg.timescale
-  const tsdbUrl = process.env.DAQ_TSDB_URL ?? `postgres://${encodeURIComponent(user)}:${encodeURIComponent(password)}@${host}:${port}/${database}`
-  // 对象存储(v2):DAQ_OS_URL 覆盖;未配置 objectstore 段 → null(磁盘降级)
-  const osCfg = cfg.objectstore
-  const objectStoreUrl = process.env.DAQ_OS_URL
-    ?? (osCfg
-      ? `http://${encodeURIComponent(osCfg.accessKey)}:${encodeURIComponent(osCfg.secretKey)}@${osCfg.host}:${osCfg.port}/${osCfg.bucket}`
-      : null)
-  return { mqttUrl, tsdbUrl, objectStoreUrl }
-}
-
-/**
- * 探测/拉起/降级判定主流程(幂等;启动插件与重连端点共用)。
- * 返回最终状态;真实后端在线时由调用方(插件)重建 queue/tsdb 单例并恢复采集。
- */
-export async function ensureDaqInfrastructure(cfg: DaqInfraConfig, projectRoot: string): Promise<DaqInfraStatus> {
-  const st = state()
-  st.cfg = cfg
-  const s = st.status
-  s.lastCheckAt = new Date().toISOString()
-  s.lastError = undefined
-
-  // 1) 直接探测配置地址
-  let probe = await waitPorts(cfg, 3000)
-  s.startedBy = 'none'
-  // 2) 不通 → 按策略 Docker 拉起
-  if (!(probe.mqtt && probe.tsdb) && cfg.startInfrastructure !== 'never') {
-    const canDocker = await dockerAvailable()
-    if (canDocker) {
-      log.info('[daq-infra] 配置地址未就绪,经 Docker Compose 拉起(daq-mosquitto + daq-timescale + daq-minio)…')
-      const up = await dockerComposeUp(projectRoot)
-      if (up.ok) {
-        probe = await waitPorts(cfg, 60_000)
-        if (probe.mqtt || probe.tsdb) s.startedBy = 'docker'
-      }
-      else {
-        s.lastError = `docker compose 拉起失败: ${up.error?.slice(0, 200)}`
-        log.error('[daq-infra]', s.lastError)
-      }
-    }
-    else if (cfg.startInfrastructure === 'auto') {
-      s.lastError = '宿主机无 Docker,直接连接配置地址'
-    }
-  }
-
-  if (probe.mqtt && probe.tsdb) {
-    if (s.startedBy === 'none') s.startedBy = 'direct'
-  }
-
-  s.mqttOnline = probe.mqtt
-  s.tsdbOnline = probe.tsdb
-  s.objectStoreOnline = probe.os
-  s.degraded = !(probe.mqtt && probe.tsdb)
-
-  if (s.degraded) {
-    const missing: string[] = []
-    if (!probe.mqtt) missing.push(`MQTT ${cfg.mqtt.host}:${cfg.mqtt.port}`)
-    if (!probe.tsdb) missing.push(`Timescale ${cfg.timescale.host}:${cfg.timescale.port}`)
-    s.warning = `数采基础设施不可达:${missing.join('、')}${s.lastError ? `(${s.lastError})` : ''}。已降级:队列→进程内 / 时序库→SQLite 仿真,在线采集已停用;修复后可重连。`
-    log.warn(`[daq-infra] 降级运行 — ${s.warning}`)
-  }
-  else {
-    s.warning = probe.os
-      ? ''
-      : `对象存储(MinIO)不可达:图像帧已降级本地磁盘(data/daq-objects/),采集不中断。`
-    log.info(`[daq-infra] 真实链路就绪(mqtt=${cfg.mqtt.host}:${cfg.mqtt.port}, timescale=${cfg.timescale.host}:${cfg.timescale.port}, objectstore=${probe.os ? `${cfg.objectstore?.host}:${cfg.objectstore?.port}` : 'disk-degraded'}, 来源=${s.startedBy})`)
-  }
-  return daqInfraStatus()
-}
-
-/** 后台自动重连(降级期间每 30s 重试;恢复后 rebuild + 恢复采集) */
-export function scheduleAutoReconnect(onRestored: () => void): void {
-  const st = state()
-  if (st.retryTimer) return
-  st.retryTimer = setInterval(() => {
-    const s = st.status
-    if (!s.degraded) return
-    void ensureDaqInfrastructure(st.cfg, st.projectRoot ?? process.cwd()).then((fresh) => {
-      if (!fresh.degraded) {
-        log.info('[daq-infra] 自动重连成功,恢复真实链路')
-        onRestored()
-      }
-    }).catch(() => { /* 下个周期再试 */ })
-  }, 30_000)
-  st.retryTimer.unref?.()
-}
+/**
+ * DAQ 基础设施编排 —— 配置驱动的真实 MQTT/Timescale 生命周期。
+ *
+ * 启动策略(config.yml daq.startInfrastructure):
+ *   auto   = 探测配置地址;端口不通且宿主有 Docker → docker compose 自动拉起,
+ *            拉起后等待就绪;无 Docker → 直连失败即降级;
+ *   always = 仅经 Docker 拉起(不探测直连);
+ *   never  = 只按配置直连。
+ *
+ * 降级语义:任一后端不可达 → 该后端回退 mock 对应物(进程内队列 / SQLite 时序仿真),
+ * 控制器停采(在线数采停用),meta 携带 warning,前端横幅 + 一键重连;
+ * 后台每 30s 自动重试,恢复后自动重启采集并切回真实后端。
+ */
+import { createLogger } from '../logger'
+import { execFile } from 'node:child_process'
+import { existsSync } from 'node:fs'
+import { createConnection } from 'node:net'
+import { promisify } from 'node:util'
+import { join } from 'node:path'
+import { composeEnvFor, credentialsEnvFilePath, loadOrCreateCredentials } from './infra-credentials'
+
+const log = createLogger('daq.infra')
+
+const execFileAsync = promisify(execFile)
+
+export interface DaqInfraConfig {
+  startInfrastructure: 'auto' | 'always' | 'never'
+  mqtt: { host: string, port: number, username?: string, password?: string, secure?: boolean, caFile?: string }
+  timescale: { host: string, port: number, user: string, password: string, database: string }
+  /** 对象存储(v2:图像帧;不可达 → 本地磁盘降级,不阻断采集) */
+  objectstore?: { host: string, port: number, accessKey: string, secretKey: string, bucket: string }
+}
+
+export interface DaqInfraStatus {
+  /** 真实 MQTT 是否在线(离线 → 进程内队列降级) */
+  mqttOnline: boolean
+  /** Timescale 是否在线(离线 → SQLite 时序仿真降级) */
+  tsdbOnline: boolean
+  /** 对象存储是否在线(离线 → 本地磁盘降级;不计入 degraded,采集不中断) */
+  objectStoreOnline: boolean
+  /** 任一降级 → 在线采集停用,前端横幅警告 */
+  degraded: boolean
+  warning: string
+  /** 基础设施拉起方式:docker / direct / none */
+  startedBy: 'docker' | 'direct' | 'none'
+  lastCheckAt: string
+  lastError?: string
+}
+
+const g = globalThis as typeof globalThis & {
+  __daqInfra?: { cfg: DaqInfraConfig, status: DaqInfraStatus, retryTimer?: NodeJS.Timeout, projectRoot?: string }
+}
+
+function state() {
+  return g.__daqInfra ??= {
+    cfg: { startInfrastructure: 'auto', mqtt: { host: '127.0.0.1', port: 1883 }, timescale: { host: '127.0.0.1', port: 5432, user: 'postgres', password: 'awshop', database: 'awshop' } },
+    status: { mqttOnline: false, tsdbOnline: false, objectStoreOnline: false, degraded: true, warning: '', startedBy: 'none', lastCheckAt: new Date().toISOString() },
+  }
+}
+
+/** TCP 端口探测(2s 超时;Docker 端口映射即代表服务可连) */
+function probePort(host: string, port: number, timeoutMs = 2000): Promise<boolean> {
+  return new Promise((resolve) => {
+    const sock = createConnection({ host, port })
+    const done = (ok: boolean): void => {
+      sock.destroy()
+      resolve(ok)
+    }
+    sock.setTimeout(timeoutMs, () => done(false))
+    sock.once('connect', () => done(true))
+    sock.once('error', () => done(false))
+  })
+}
+
+async function dockerAvailable(): Promise<boolean> {
+  try {
+    await execFileAsync('docker', ['--version'], { timeout: 5000 })
+    return true
+  }
+  catch {
+    return false
+  }
+}
+
+async function dockerComposeUp(projectRoot: string): Promise<{ ok: boolean, error?: string }> {
+  try {
+    // 凭据由服务端生成/采纳后经环境注入(仓库零口令字面量,audit P0-5);
+    // 已落盘时额外给 --env-file,便于 compose 在子进程 env 被裁剪时仍能解析。
+    const envFile = credentialsEnvFilePath()
+    const args = ['compose', '-f', join(projectRoot, 'docker-compose.yml'), 'up', '-d', 'daq-mosquitto', 'daq-timescale', 'daq-minio']
+    const finalArgs = existsSync(envFile) ? ['compose', '--env-file', envFile, ...args.slice(1)] : args
+    await execFileAsync('docker', finalArgs, { timeout: 180_000, cwd: projectRoot, env: composeEnvFor() })
+    return { ok: true }
+  }
+  catch (err) {
+    return { ok: false, error: err instanceof Error ? err.message : String(err) }
+  }
+}
+
+/** 等待三个端口就绪(拉起后服务初始化需要数秒;最长 waitMs) */
+async function waitPorts(cfg: DaqInfraConfig, waitMs: number): Promise<{ mqtt: boolean, tsdb: boolean, os: boolean }> {
+  const deadline = Date.now() + waitMs
+  let mqtt = false
+  let tsdb = false
+  let os = false
+  while (Date.now() < deadline && !(mqtt && tsdb && os)) {
+    mqtt = mqtt || await probePort(cfg.mqtt.host, cfg.mqtt.port, 1200)
+    tsdb = tsdb || await probePort(cfg.timescale.host, cfg.timescale.port, 1200)
+    os = os || await probePort(cfg.objectstore?.host ?? '127.0.0.1', cfg.objectstore?.port ?? 9000, 1200)
+    if (!(mqtt && tsdb && os)) await new Promise(r => setTimeout(r, 1500))
+  }
+  return { mqtt, tsdb, os }
+}
+
+export function daqInfraStatus(): DaqInfraStatus {
+  return { ...state().status }
+}
+
+/**
+ * 后端装配失败(S10):端口探测通过但 TSDB/队列/对象存储 init 失败时,
+ * 必须把状态显式置为 degraded —— 否则 30s 自动重连永不触发,采样会写进占位后端。
+ */
+export function markDaqInfraFailure(err: unknown): void {
+  const s = state().status
+  s.degraded = true
+  s.lastCheckAt = new Date().toISOString()
+  s.lastError = err instanceof Error ? err.message : String(err)
+  s.warning = `数采后端初始化失败,已降级并停采:${s.lastError}`
+  log.error('[daq-infra]', s.warning)
+}
+
+/** 运行配置快照(启动插件装配时写入;重连端点复用) */
+export function daqInfraConfig(): DaqInfraConfig {
+  return state().cfg
+}
+
+/** 配置 → 连接 URL(env 显式覆盖优先,便于外部托管 broker/db/对象存储;S1:支持凭据/TLS) */
+export function daqUrls(cfg = state().cfg): { mqttUrl: string | null, tsdbUrl: string | null, objectStoreUrl: string | null } {
+  // S1:凭据/TLS 优先 env(DAQ_MQTT_URL 整串覆盖),否则由 config.mqtt 拼(含 user:pass@ 与 mqtts://)
+  let mqttUrl: string
+  if (process.env.DAQ_MQTT_URL) {
+    mqttUrl = process.env.DAQ_MQTT_URL
+  }
+  else {
+    const scheme = cfg.mqtt.secure ? 'mqtts' : 'mqtt'
+    const cred = cfg.mqtt.username
+      ? `${encodeURIComponent(cfg.mqtt.username)}:${encodeURIComponent(cfg.mqtt.password ?? '')}@`
+      : ''
+    mqttUrl = `${scheme}://${cred}${cfg.mqtt.host}:${cfg.mqtt.port}`
+  }
+  // 配置里的凭据留空 = 用本机生成/采纳的基础设施凭据(与 docker compose 同一份;
+  // 见 infra-credentials.ts)。runtimeConfig 的装配早于配置服务 init,不能只依赖它。
+  const local = (() => {
+    try {
+      return loadOrCreateCredentials({ adopt: false })
+    }
+    catch {
+      return null
+    }
+  })()
+  const { user, password, database, host, port } = cfg.timescale
+  // 优先级:显式环境变量(外部托管) > config.yml > 本机凭据文件(compose 同一份)
+  const tsPassword = process.env.AW_DAQ_TIMESCALE_PASSWORD || password || local?.tsSecret || ''
+  const tsdbUrl = process.env.DAQ_TSDB_URL ?? `postgres://${encodeURIComponent(user)}:${encodeURIComponent(tsPassword)}@${host}:${port}/${database}`
+  // 对象存储(v2):DAQ_OS_URL 覆盖;未配置 objectstore 段 → null(磁盘降级)
+  const osCfg = cfg.objectstore
+  const osAccess = process.env.AW_DAQ_OBJECTSTORE_ACCESSKEY || osCfg?.accessKey || local?.osAccessId || ''
+  const osSecret = process.env.AW_DAQ_OBJECTSTORE_SECRETKEY || osCfg?.secretKey || local?.osSecret || ''
+  const objectStoreUrl = process.env.DAQ_OS_URL
+    ?? (osCfg
+      ? `http://${encodeURIComponent(osAccess)}:${encodeURIComponent(osSecret)}@${osCfg.host}:${osCfg.port}/${osCfg.bucket}`
+      : null)
+  return { mqttUrl, tsdbUrl, objectStoreUrl }
+}
+
+/**
+ * 探测/拉起/降级判定主流程(幂等;启动插件与重连端点共用)。
+ * 返回最终状态;真实后端在线时由调用方(插件)重建 queue/tsdb 单例并恢复采集。
+ */
+export async function ensureDaqInfrastructure(cfg: DaqInfraConfig, projectRoot: string): Promise<DaqInfraStatus> {
+  const st = state()
+  st.cfg = cfg
+  const s = st.status
+  s.lastCheckAt = new Date().toISOString()
+  s.lastError = undefined
+
+  // 1) 直接探测配置地址
+  let probe = await waitPorts(cfg, 3000)
+  s.startedBy = 'none'
+  // 2) 不通 → 按策略 Docker 拉起
+  if (!(probe.mqtt && probe.tsdb) && cfg.startInfrastructure !== 'never') {
+    const canDocker = await dockerAvailable()
+    if (canDocker) {
+      log.info('[daq-infra] 配置地址未就绪,经 Docker Compose 拉起(daq-mosquitto + daq-timescale + daq-minio)…')
+      const up = await dockerComposeUp(projectRoot)
+      if (up.ok) {
+        probe = await waitPorts(cfg, 60_000)
+        if (probe.mqtt || probe.tsdb) s.startedBy = 'docker'
+      }
+      else {
+        s.lastError = `docker compose 拉起失败: ${up.error?.slice(0, 200)}`
+        log.error('[daq-infra]', s.lastError)
+      }
+    }
+    else if (cfg.startInfrastructure === 'auto') {
+      s.lastError = '宿主机无 Docker,直接连接配置地址'
+    }
+  }
+
+  if (probe.mqtt && probe.tsdb) {
+    if (s.startedBy === 'none') s.startedBy = 'direct'
+  }
+
+  s.mqttOnline = probe.mqtt
+  s.tsdbOnline = probe.tsdb
+  s.objectStoreOnline = probe.os
+  s.degraded = !(probe.mqtt && probe.tsdb)
+
+  if (s.degraded) {
+    const missing: string[] = []
+    if (!probe.mqtt) missing.push(`MQTT ${cfg.mqtt.host}:${cfg.mqtt.port}`)
+    if (!probe.tsdb) missing.push(`Timescale ${cfg.timescale.host}:${cfg.timescale.port}`)
+    s.warning = `数采基础设施不可达:${missing.join('、')}${s.lastError ? `(${s.lastError})` : ''}。已降级:队列→进程内 / 时序库→SQLite 仿真,在线采集已停用;修复后可重连。`
+    log.warn(`[daq-infra] 降级运行 — ${s.warning}`)
+  }
+  else {
+    s.warning = probe.os
+      ? ''
+      : `对象存储(MinIO)不可达:图像帧已降级本地磁盘(data/daq-objects/),采集不中断。`
+    log.info(`[daq-infra] 真实链路就绪(mqtt=${cfg.mqtt.host}:${cfg.mqtt.port}, timescale=${cfg.timescale.host}:${cfg.timescale.port}, objectstore=${probe.os ? `${cfg.objectstore?.host}:${cfg.objectstore?.port}` : 'disk-degraded'}, 来源=${s.startedBy})`)
+  }
+  return daqInfraStatus()
+}
+
+/** 后台自动重连(降级期间每 30s 重试;恢复后 rebuild + 恢复采集) */
+export function scheduleAutoReconnect(onRestored: () => void): void {
+  const st = state()
+  if (st.retryTimer) return
+  st.retryTimer = setInterval(() => {
+    const s = st.status
+    if (!s.degraded) return
+    void ensureDaqInfrastructure(st.cfg, st.projectRoot ?? process.cwd()).then((fresh) => {
+      if (!fresh.degraded) {
+        log.info('[daq-infra] 自动重连成功,恢复真实链路')
+        onRestored()
+      }
+    }).catch(() => { /* 下个周期再试 */ })
+  }, 30_000)
+  st.retryTimer.unref?.()
+}