agentworkshop 0.7.54 → 0.7.59

package/server/plugins/backup.ts

@@ -6,7 +6,7 @@
  * setInterval + globalThis key 防 HMR 重复 + unref()。dev 与生产均生效
  * (备份无副作用,始终开启;BACKUP_DISABLED=1 可关)。
  */
-import { existsSync, mkdirSync, readdirSync, renameSync, rmSync } from 'node:fs'
+import { existsSync, mkdirSync, readdirSync, renameSync, rmSync, statSync, cpSync, writeFileSync } from 'node:fs'
 import { copyFile as copyFileAsync } from 'node:fs/promises'
 import { resolve } from 'node:path'
 import { ensureDataDir } from '@/shared/config/home.mjs'
@@ -17,6 +17,68 @@ const g = globalThis as typeof globalThis & { __awBackupTimer?: NodeJS.Timeout,
 
 const DB_FILES = ['workshop.sqlite', 'users.sqlite', 'daq-timeseries.sqlite'] as const
 
+/** 目录树字节总量(快速 walk;仅 size 求和,用于备份上限判定) */
+function dirBytes(dir: string): number {
+  let total = 0
+  let entries: Array<import('node:fs').Dirent>
+  try {
+    entries = readdirSync(dir, { withFileTypes: true })
+  }
+  catch {
+    return 0
+  }
+  for (const e of entries) {
+    const p = resolve(dir, e.name)
+    if (e.isDirectory()) total += dirBytes(p)
+    else if (e.isFile()) {
+      try {
+        total += statSync(p).size
+      }
+      catch { /* 竞态:文件恰好消失,按 0 计 */ }
+    }
+  }
+  return total
+}
+
+/**
+ * 对象面备份(daq-objects 本地对象存储 + daq-exports 导出宽表)。
+ * 2026-10-09 生产化扩围:此前只备三库 + JSON 面,图像帧对象/导出数据不在备份
+ * 范围(真实产线丢单对象=丢检验证据)。带体量上限(AW_BACKUP_OBJECTS_MAX_MB,
+ * 缺省 200MB):超限跳过并显式落日志/账,提示走 infra 级卷备份,不做半份拷贝。
+ */
+async function backupObjects(dataDir: string, backupDir: string, stamp: string, manifest: Record<string, unknown>): Promise<void> {
+  const capMb = Math.max(0, Number(process.env.AW_BACKUP_OBJECTS_MAX_MB ?? 200))
+  if (capMb === 0) {
+    manifest.objects = { skipped: 'disabled(env AW_BACKUP_OBJECTS_MAX_MB=0)' }
+    return
+  }
+  const bundleDir = resolve(backupDir, `objects-${stamp}`)
+  const included: string[] = []
+  const skipped: string[] = []
+  mkdirSync(bundleDir, { recursive: true })
+  for (const dirName of ['daq-objects', 'daq-exports']) {
+    const src = resolve(dataDir, dirName)
+    if (!existsSync(src)) continue
+    const bytes = dirBytes(src)
+    if (bytes > capMb * 1024 * 1024) {
+      skipped.push(`${dirName}(${Math.round(bytes / 1024 / 1024)}MB > 上限 ${capMb}MB)`)
+      continue
+    }
+    try {
+      cpSync(src, resolve(bundleDir, dirName), { recursive: true })
+      included.push(`${dirName}(${Math.round(bytes / 1024 / 1024)}MB)`)
+    }
+    catch (err) {
+      skipped.push(`${dirName}(拷贝失败: ${err instanceof Error ? err.message : String(err)})`)
+    }
+  }
+  manifest.objects = { bundle: `objects-${stamp}`, included, skipped }
+  if (skipped.length) console.warn('[backup] 对象面部分跳过:', skipped.join('; '))
+}
+
+/** TIMESCALE/MINIO 卷说明(docker-compose 部署形态;文件级备份覆盖不到,RESTORE 文档写明) */
+const INFRA_VOLUMES_NOTE = 'TimescaleDB/MinIO 卷不在文件备份范围(docker 部署用 pg_dump/mc mirror 或卷快照)'
+
 /**
  * 单库在线快照(hardening ST-2 终版:零侵入文件拷贝)。
  *
@@ -91,12 +153,14 @@ export async function backupOnce(dataDir: string): Promise<void> {
   const backupDir = resolve(dataDir, 'backups')
   mkdirSync(backupDir, { recursive: true })
   const stamp = new Date().toISOString().replace(/[:.]/g, '-').slice(0, 19)
+  const done: string[] = []
   for (const file of DB_FILES) {
     const src = resolve(dataDir, file)
     if (!existsSync(src) || !backupRegistry.has(src)) continue
     const target = resolve(backupDir, `${file}.${stamp}.bak`)
     try {
       await backupOne(src, target)
+      done.push(file)
       console.log('[backup]', file, '快照完成(serialize)')
     }
     catch (err) {
@@ -105,6 +169,13 @@ export async function backupOnce(dataDir: string): Promise<void> {
   }
   // JSON 仓储 + 配置面快照(三库之外的可重建性半边)
   await backupFlatFiles(dataDir, backupDir, stamp)
+  // 对象面(图像帧对象存储 + 导出宽表;带体量上限,超限显式跳过)+ MANIFEST
+  const manifest: Record<string, unknown> = { stamp, dbs: done, infraNote: INFRA_VOLUMES_NOTE }
+  await backupObjects(dataDir, backupDir, stamp, manifest)
+  try {
+    writeFileSync(resolve(backupDir, `manifest-${stamp}.json`), JSON.stringify(manifest, null, 2))
+  }
+  catch { /* 清单写失败不影响备份本体 */ }
   // 轮转:每库仅保留最近 backup.keep 份(按文件名内时间戳倒序;env BACKUP_KEEP 兼容)
   const keep = Math.max(1, backupSettings().keep)
   for (const file of DB_FILES) {
@@ -119,17 +190,29 @@ export async function backupOnce(dataDir: string): Promise<void> {
       catch { /* 轮转失败不致命 */ }
     }
   }
-  // files-<stamp>/ 目录同 keep 轮转
-  const fileBundles = readdirSync(backupDir)
-    .filter(f => f.startsWith('files-') && !f.includes('.'))
-    .sort()
-    .reverse()
-  for (const stale of fileBundles.slice(keep)) {
-    try {
-      rmSync(resolve(backupDir, stale), { recursive: true })
+  // files-<stamp>/ 与 objects-<stamp>/、manifest-*.json 同 keep 轮转
+  for (const prefix of ['files-', 'objects-']) {
+    const bundles = readdirSync(backupDir)
+      .filter(f => f.startsWith(prefix) && !f.includes('.'))
+      .sort()
+      .reverse()
+    for (const stale of bundles.slice(keep)) {
+      try {
+        rmSync(resolve(backupDir, stale), { recursive: true })
+      }
+      catch { /* 轮转失败不致命 */ }
+    }
+  }
+  try {
+    const mans = readdirSync(backupDir).filter(f => f.startsWith('manifest-') && f.endsWith('.json')).sort().reverse()
+    for (const stale of mans.slice(keep)) {
+      try {
+        rmSync(resolve(backupDir, stale))
+      }
+      catch { /* ignore */ }
     }
-    catch { /* 轮转失败不致命 */ }
   }
+  catch { /* ignore */ }
   console.log('[backup] 快照完成 →', backupDir)
   // R4:记录最近一次成功备份时间,供 /api/metrics 观测
   g.__awBackupLastAt = new Date().toISOString()
@@ -138,7 +221,7 @@ export async function backupOnce(dataDir: string): Promise<void> {
     recordOps({
       actor: 'system', actorName: 'backup', actorKind: 'system',
       action: 'system.backup.run', kind: 'system',
-      summary: `数据库快照完成(${DB_FILES.length} 库 + JSON/配置)→ ${backupDir}`,
+      summary: `数据库快照完成(${done.length} 库 + JSON/配置 + 对象面)→ ${backupDir}`,
     })
   }).catch(() => {})
 }

package/server/plugins/daq.ts

@@ -9,6 +9,7 @@ import { useRuntimeConfig } from '#imports'
 import { daqUrls, ensureDaqInfrastructure, scheduleAutoReconnect, type DaqInfraConfig } from '../services/workshop/daq/infra'
 import { rebuildTsdb } from '../services/workshop/daq/storage'
 import { rebuildObjectStore } from '../services/workshop/daq/objectstore'
+import { startDaqObjectGc } from '../services/workshop/daq/objectstore/gc'
 import { rebuildDaqQueue } from '../services/workshop/daq/bus'
 import { getDaqController } from '../services/workshop/daq/daq-controller'
 
@@ -71,6 +72,8 @@ export default async function daqPlugin() {
   scheduleAutoReconnect(() => {
     void applyInfra(cfg).catch(() => {})
   })
+  // P0-2 对象存储 GC:帧行出清后像素文件孤儿回收(90s 首轮 + 24h 周期;disk 后端)
+  startDaqObjectGc()
   // 重连端点复用同一装配函数
   ;(globalThis as unknown as { __daqApplyInfra?: (c: DaqInfraConfig) => Promise<void> }).__daqApplyInfra = applyInfra
 }

package/server/plugins/memory-watch.ts

@@ -0,0 +1,92 @@
+/**
+ * 内存水位监控(P0-4)—— 5 小时级慢泄漏 OOM(exit 134)的止血观测面。
+ *
+ * 30s 采样 heapUsed/rss/external;>75% warn(带较上次增量)、>90% error;
+ * 采样环形序列周期落盘 <configRoot>/.runtime/last-mem.json —— 重启自愈后
+ * 泄漏差分分析直接读该文件(0h vs Nh 增长对比),无需再挂探针压测。
+ *
+ * 限额取 v8.getHeapStatistics().heap_size_limit(实际生效的 --max-old-space-size,
+ * 含 NODE_OPTIONS 传入);globalThis 守卫防 HMR 重复挂表,unref 不阻退出。
+ */
+import { mkdirSync, writeFileSync } from 'node:fs'
+import { join } from 'node:path'
+import v8 from 'node:v8'
+import { ensureDataDir } from '@/shared/config/home.mjs'
+
+interface MemSample { at: string, heapUsedMb: number, rssMb: number, externalMb: number, ratio: number }
+
+const SAMPLE_MS = 30_000
+const RING_CAP = 120 // 1h 窗口
+const PERSIST_EVERY = 10 // 每 10 拍(5min)落盘一次
+const WARN_RATIO = 0.75
+const ERROR_RATIO = 0.9
+
+const g = globalThis as typeof globalThis & {
+  __awMemWatchTimer?: NodeJS.Timeout
+  __awMemRing?: MemSample[]
+}
+
+function limitBytes(): number {
+  try {
+    return v8.getHeapStatistics().heap_size_limit
+  }
+  catch {
+    return 0
+  }
+}
+
+export function memorySnapshot(): { heapUsedMb: number, heapLimitMb: number, rssMb: number, externalMb: number, ratio: number } {
+  const mu = process.memoryUsage()
+  const limit = limitBytes()
+  const heapLimitMb = Math.round(limit / 1048576)
+  const heapUsedMb = Math.round(mu.heapUsed / 1048576)
+  return {
+    heapUsedMb,
+    heapLimitMb,
+    rssMb: Math.round(mu.rss / 1048576),
+    externalMb: Math.round(mu.external / 1048576),
+    ratio: limit > 0 ? Number((mu.heapUsed / limit).toFixed(4)) : 0,
+  }
+}
+
+function persist(ring: MemSample[]): void {
+  try {
+    const dir = join(ensureDataDir(), '..', '.runtime')
+    mkdirSync(dir, { recursive: true })
+    writeFileSync(join(dir, 'last-mem.json'), JSON.stringify({ limitMb: Math.round(limitBytes() / 1048576), samples: ring.slice(-RING_CAP) }, null, 2))
+  }
+  catch { /* 落盘失败不影响观测 */ }
+}
+
+export default function memoryWatchPlugin() {
+  if (g.__awMemWatchTimer) return
+  const ring: MemSample[] = g.__awMemRing ??= []
+  let lastHeapUsedMb = 0
+  let ticks = 0
+
+  const sample = (): void => {
+    const snap = memorySnapshot()
+    const rec: MemSample = {
+      at: new Date().toISOString(),
+      heapUsedMb: snap.heapUsedMb,
+      rssMb: snap.rssMb,
+      externalMb: snap.externalMb,
+      ratio: snap.ratio,
+    }
+    ring.push(rec)
+    if (ring.length > RING_CAP) ring.shift()
+    const deltaMb = rec.heapUsedMb - lastHeapUsedMb
+    lastHeapUsedMb = rec.heapUsedMb
+    if (rec.ratio >= ERROR_RATIO) {
+      console.error(`[memory-watch] 堆水位 ${Math.round(rec.ratio * 100)}%(heap ${rec.heapUsedMb}/${Math.round(limitBytes() / 1048576)}MB,rss ${rec.rssMb}MB,${deltaMb >= 0 ? '+' : ''}${deltaMb}MB/30s)—— 逼近 OOM,观察自动重启`)
+    }
+    else if (rec.ratio >= WARN_RATIO) {
+      console.warn(`[memory-watch] 堆水位 ${Math.round(rec.ratio * 100)}%(heap ${rec.heapUsedMb}MB,${deltaMb >= 0 ? '+' : ''}${deltaMb}MB/30s)`)
+    }
+    if (++ticks % PERSIST_EVERY === 0) persist(ring)
+  }
+
+  sample()
+  g.__awMemWatchTimer = setInterval(sample, SAMPLE_MS)
+  g.__awMemWatchTimer.unref?.()
+}

package/server/plugins/parent-watch.ts

@@ -0,0 +1,22 @@
+/**
+ * 父进程看门(P0-4 respawn 配套)—— 生产由 scripts/start.mjs 以子进程方式拉起服务,
+ * 子进程必须先于一切感知父进程死亡并自退:父被硬杀(taskkill 不带 /T)后孤儿子进程
+ * 会占住端口,新实例被端口顺延推到 +1(双实例假象)。
+ * env AW_PARENT_PID 未设(手动 node .output 直启)时不启用。
+ */
+const g = globalThis as typeof globalThis & { __awParentWatchTimer?: NodeJS.Timeout }
+
+export default function parentWatchPlugin() {
+  const ppid = Number(process.env.AW_PARENT_PID ?? 0)
+  if (!Number.isInteger(ppid) || ppid <= 0 || g.__awParentWatchTimer) return
+  g.__awParentWatchTimer = setInterval(() => {
+    try {
+      process.kill(ppid, 0)
+    }
+    catch {
+      console.error('[parent-watch] 父进程已退出 —— 子进程自退(防孤儿占端口)')
+      process.exit(1)
+    }
+  }, 5000)
+  g.__awParentWatchTimer.unref?.()
+}