Dev

.github/workflows/docker-publish.yml

@@ -115,9 +115,10 @@ jobs:
           file: ./Dockerfile
           platforms: ${{ matrix.platform }}
           push: true
-          # 使用带架构后缀的临时标签
+          # 同时推送到两个 Registry 的临时架构标签，确保层已就绪，避免合并时跨 Registry 复制层
           tags: |
             ${{ env.REGISTRY }}/${{ github.repository_owner }}/${{ env.IMAGE_NAME }}:build-${{ github.run_id }}-${{ matrix.suffix }}
+            ${{ (github.event_name != 'pull_request' && secrets.DOCKERHUB_TOKEN != '') && format('{0}:build-{1}-{2}', env.DOCKERHUB_REPO, github.run_id, matrix.suffix) || '' }}
           labels: ${{ steps.meta.outputs.labels }}
           cache-from: type=gha,scope=${{ matrix.platform }}
           cache-to: type=gha,mode=max,scope=${{ matrix.platform }}
@@ -198,10 +199,30 @@ jobs:
           TAGS="${{ steps.meta.outputs.tags }}"
           echo "$TAGS" | while IFS= read -r TAG; do
             [ -z "$TAG" ] && continue
-            echo "Creating manifest for: $TAG"
-            docker buildx imagetools create -t "$TAG" \
-              "${{ env.REGISTRY }}/${{ github.repository_owner }}/${{ env.IMAGE_NAME }}@${AMD64_DIGEST}" \
-              "${{ env.REGISTRY }}/${{ github.repository_owner }}/${{ env.IMAGE_NAME }}@${ARM64_DIGEST}"
+
+            # 确定当前标签所属的 Registry 以选择同源的源镜像进行合并（避免跨 Registry 复制 Blob 导致 400 错误）
+            if [[ "$TAG" == "${{ env.REGISTRY }}"* ]]; then
+               SRC_PREFIX="${{ env.REGISTRY }}/${{ github.repository_owner }}/${{ env.IMAGE_NAME }}"
+            else
+               SRC_PREFIX="${{ env.DOCKERHUB_REPO }}"
+            fi
+
+            echo "Creating manifest for: $TAG using source: $SRC_PREFIX"
+
+            # 增加重试机制，应对 Registry 极端情况下的抖动
+            for i in {1..3}; do
+              docker buildx imagetools create -t "$TAG" \
+                "${SRC_PREFIX}@${AMD64_DIGEST}" \
+                "${SRC_PREFIX}@${ARM64_DIGEST}" && break || {
+                if [ $i -lt 3 ]; then
+                  echo "Push failed, retrying in 10s ($i/3)..."
+                  sleep 10
+                else
+                  echo "Failed to create manifest for $TAG after 3 attempts."
+                  exit 1
+                fi
+              }
+            done
           done
 
       - name: Image digest

Dockerfile

@@ -96,6 +96,7 @@ COPY --from=agent-builder --chown=nodejs:nodejs /app/agent-go/agent-windows-amd6
 # 4. 复制后端源码 (不包含 node_modules)
 COPY --chown=nodejs:nodejs server.js ./
 COPY --chown=nodejs:nodejs src ./src
+COPY --chown=nodejs:nodejs modules ./modules
 
 ENV NODE_ENV=production \
     PORT=3000 \

agent-go/collector.go

@@ -100,17 +100,29 @@ type Collector struct {
 	lastGPUPower   float64
 	lastGPUTime    time.Time
 
-	// CPU 采集缓存 (保持上次有效值，避免返回 0)
-	lastCPUUsage float64
+	// GPU 采集频率控制
+	lastGPUMetadataTime time.Time
+
+	// CPU 采集缓存
 	lastCPUTime  time.Time
+	lastCPUUsage float64
+
+	// Windows Native (PDH)
+	pdhQuery   uintptr
+	pdhCounter uintptr
+
+	// NVIDIA Native (NVML)
+	nvmlLib         any
+	nvmlInitialized bool
 }
 
 // NewCollector 创建采集器
 func NewCollector() *Collector {
 	return &Collector{
-		lastNetTime: time.Now(),
-		lastGPUTime: time.Now().Add(-1 * time.Hour), // 确保第一次采集立即执行
-		lastCPUTime: time.Now().Add(-1 * time.Hour), // 确保第一次采集立即执行
+		lastNetTime:         time.Now(),
+		lastGPUTime:         time.Now().Add(-1 * time.Hour), // 确保第一次采集立即执行
+		lastCPUTime:         time.Now().Add(-1 * time.Hour), // 确保第一次采集立即执行
+		lastGPUMetadataTime: time.Now().Add(-1 * time.Hour), // 确保第一次采集立即执行
 	}
 }
 
@@ -191,6 +203,7 @@ func (c *Collector) CollectHostInfo() *HostInfo {
 	gpuModels, gpuMemTotal := c.collectGPUMetadata()
 	info.GPU = gpuModels
 	info.GPUMemTotal = gpuMemTotal
+	c.lastGPUMetadataTime = time.Now()
 
 	c.cachedHostInfo = info
 	return info
@@ -321,21 +334,27 @@ func (c *Collector) CollectState() *State {
 		c.lastGPUTime = time.Now()
 	}
 
-	// 补救措施：如果显存总量为 0，尝试重新获取静态信息
+	// 补救措施：如果显存总量为 0，尝试重新获取静态信息 (增加冷却时间，防止频繁调用 PowerShell)
 	if c.cachedHostInfo != nil && c.cachedHostInfo.GPUMemTotal == 0 {
-		go func() {
-			c.mu.Lock()
-			defer c.mu.Unlock()
-			// 再次检查，防止并发重复
-			if c.cachedHostInfo.GPUMemTotal == 0 {
+		c.mu.Lock()
+		shouldRetry := time.Since(c.lastGPUMetadataTime) > 10*time.Minute
+		if shouldRetry {
+			c.lastGPUMetadataTime = time.Now() // 预设时间，防止下一秒再次触发
+		}
+		c.mu.Unlock()
+
+		if shouldRetry {
+			go func() {
 				models, total := c.collectGPUMetadata()
 				if total > 0 {
+					c.mu.Lock()
 					c.cachedHostInfo.GPU = models
 					c.cachedHostInfo.GPUMemTotal = total
+					c.mu.Unlock()
 					fmt.Printf("[Collector] GPU metadata refreshed: %d MiB\n", total/1024/1024)
 				}
-			}
-		}()
+			}()
+		}
 	}
 	state.GPU = c.lastGPUUsage
 	state.GPUMemUsed = c.lastGPUMemUsed
@@ -545,6 +564,14 @@ func (c *Collector) collectGPUState() (float64, uint64, float64) {
 
 	// 2. 如果没有 NVIDIA，尝试其他方案
 	if runtime.GOOS == "windows" {
+		// 如果 meta 数据中已经确认没有 GPU 型号，则不再尝试采集使用率，避免频繁调用 PowerShell
+		c.mu.Lock()
+		hasGPU := c.cachedHostInfo != nil && len(c.cachedHostInfo.GPU) > 0
+		c.mu.Unlock()
+		if !hasGPU {
+			return 0, 0, 0
+		}
+
 		// Windows: 使用 Performance Counter 采集所有 GPU
 		return c.collectGPUStateWindows()
 	} else if runtime.GOOS == "linux" {
@@ -555,8 +582,14 @@ func (c *Collector) collectGPUState() (float64, uint64, float64) {
 	return 0, 0, 0
 }
 
-// collectNvidiaGPUState 使用 nvidia-smi 采集 NVIDIA GPU 状态
+// collectNvidiaGPUState 使用 NVML (优先) 或 nvidia-smi 采集 NVIDIA GPU 状态
 func (c *Collector) collectNvidiaGPUState(nvidiaSmi string) (float64, uint64, float64) {
+	// 1. 尝试使用原生 NVML API (性能更高，不产生新进程)
+	if usage, usedMem, power, ok := c.collectNvidiaGPUStateNative(); ok {
+		return usage, usedMem, power
+	}
+
+	// 2. 回退到 nvidia-smi 命令行工具
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
 	defer cancel()
 
@@ -597,14 +630,19 @@ func (c *Collector) collectNvidiaGPUState(nvidiaSmi string) (float64, uint64, fl
 	return totalUsage / float64(count), totalUsedMem, totalPower
 }
 
-// collectGPUStateWindows Windows 下采集 AMD/Intel GPU 使用率
-// 使用 PowerShell 查询 GPU Engine 性能计数器
+// collectGPUStateWindows Windows 下采集 AMD/Intel/NVIDIA GPU 使用率
+// 优先使用 PDH 性能计数器 API，回退到 PowerShell
 func (c *Collector) collectGPUStateWindows() (float64, uint64, float64) {
+	// 1. 尝试使用原生 PDH API (性能极高，无额外进程)
+	if usage, ok := c.collectGPUUsagePDH(); ok {
+		return usage, 0, 0
+	}
+
+	// 2. 回退到 PowerShell (仅在 PDH 失败时使用)
 	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
 	defer cancel()
 
 	// 方案1: 使用 GPU Engine 性能计数器 (Windows 10 1709+)
-	// 查询所有 GPU 3D 引擎的使用率
 	psCmd := `
 $counters = Get-Counter '\GPU Engine(*engtype_3D)\Utilization Percentage' -ErrorAction SilentlyContinue
 if ($counters) {
@@ -624,9 +662,6 @@ if ($counters) {
 	}
 
 	usage, _ := strconv.ParseFloat(strings.TrimSpace(string(output)), 64)
-	if usage > 0 {
-		// fmt.Printf("[Collector] GPU (Win PerfCounter): %.1f%%\n", usage)
-	}
 	return usage, 0, 0
 }
 
@@ -802,6 +837,7 @@ func (c *Collector) collectIntelGPULinux() float64 {
 	return 0
 }
 
+
 func (c *Collector) getNvidiaSmiPath() string {
 	if runtime.GOOS == "windows" {
 		possiblePaths := []string{

agent-go/collector_unix.go

@@ -0,0 +1,14 @@
+//go:build !windows
+
+package main
+
+// collectGPUUsagePDH Windows-only stub
+func (c *Collector) collectGPUUsagePDH() (float64, bool) {
+	return 0, false
+}
+
+// collectNvidiaGPUStateNative Non-Windows stub
+// (On Linux it currently falls back to nvidia-smi command line)
+func (c *Collector) collectNvidiaGPUStateNative() (float64, uint64, float64, bool) {
+	return 0, 0, 0, false
+}

agent-go/collector_windows.go

@@ -0,0 +1,137 @@
+//go:build windows
+
+package main
+
+import (
+	"runtime"
+	"syscall"
+	"unsafe"
+)
+
+var (
+	modPdh                          = syscall.NewLazyDLL("pdh.dll")
+	procPdhOpenQuery                = modPdh.NewProc("PdhOpenQueryW")
+	procPdhAddEnglishCounter        = modPdh.NewProc("PdhAddEnglishCounterW")
+	procPdhCollectQueryData         = modPdh.NewProc("PdhCollectQueryData")
+	procPdhGetFormattedCounterValue = modPdh.NewProc("PdhGetFormattedCounterValue")
+	procPdhCloseQuery               = modPdh.NewProc("PdhCloseQuery")
+)
+
+type pdh_fmt_countervalue_double struct {
+	CStatus     uint32
+	DummyStruct [4]byte // padding for 64-bit alignment
+	DoubleValue float64
+}
+
+// collectGPUUsagePDH 使用原生 PDH API 获取所有 GPU 的 3D 引擎平均使用率
+func (c *Collector) collectGPUUsagePDH() (float64, bool) {
+	if runtime.GOOS != "windows" {
+		return 0, false
+	}
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	// 初始化查询
+	if c.pdhQuery == 0 {
+		var query uintptr
+		ret, _, _ := procPdhOpenQuery.Call(0, 0, uintptr(unsafe.Pointer(&query)))
+		if ret != 0 {
+			return 0, false
+		}
+		c.pdhQuery = query
+
+		// 添加计数器 (使用通配符获取所有 GPU 的 3D 引擎使用率)
+		// 使用 English 名称确保兼容性
+		counterPath := "\\GPU Engine(*engtype_3D)\\Utilization Percentage"
+		pathPtr, _ := syscall.UTF16PtrFromString(counterPath)
+		var counter uintptr
+		ret, _, _ = procPdhAddEnglishCounter.Call(c.pdhQuery, uintptr(unsafe.Pointer(pathPtr)), 0, uintptr(unsafe.Pointer(&counter)))
+		if ret != 0 {
+			procPdhCloseQuery.Call(c.pdhQuery)
+			c.pdhQuery = 0
+			return 0, false
+		}
+		c.pdhCounter = counter
+
+		// 第一次采集建立基准
+		procPdhCollectQueryData.Call(c.pdhQuery)
+		return 0, true
+	}
+
+	// 执行采集
+	ret, _, _ := procPdhCollectQueryData.Call(c.pdhQuery)
+	if ret != 0 {
+		return 0, false
+	}
+
+	// 获取格式化后的值
+	var value pdh_fmt_countervalue_double
+	const PDH_FMT_DOUBLE = 0x00000200
+	ret, _, _ = procPdhGetFormattedCounterValue.Call(c.pdhCounter, PDH_FMT_DOUBLE, 0, uintptr(unsafe.Pointer(&value)))
+	if ret != 0 {
+		return 0, false
+	}
+
+	return value.DoubleValue, true
+}
+
+// NVIDIA NVML 原生支持 (Windows 版)
+func (c *Collector) collectNvidiaGPUStateNative() (float64, uint64, float64, bool) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	if !c.nvmlInitialized {
+		if c.nvmlLib == nil {
+			c.nvmlLib = syscall.NewLazyDLL("nvml.dll")
+		}
+
+		lib := c.nvmlLib.(*syscall.LazyDLL)
+		// 尝试初始化
+		initProc := lib.NewProc("nvmlInit_v2")
+		if err := initProc.Find(); err != nil {
+			return 0, 0, 0, false
+		}
+		ret, _, _ := initProc.Call()
+		if ret != 0 {
+			return 0, 0, 0, false
+		}
+		c.nvmlInitialized = true
+	}
+
+	lib := c.nvmlLib.(*syscall.LazyDLL)
+	// 获取第一个设备的句柄 (简化处理)
+	getHandle := lib.NewProc("nvmlDeviceGetHandleByIndex_v2")
+	var device uintptr
+	ret, _, _ := getHandle.Call(0, uintptr(unsafe.Pointer(&device)))
+	if ret != 0 {
+		return 0, 0, 0, false
+	}
+
+	// 获取利用率
+	getUtil := lib.NewProc("nvmlDeviceGetUtilizationRates")
+	var util struct {
+		GPU    uint32
+		Memory uint32
+	}
+	ret, _, _ = getUtil.Call(device, uintptr(unsafe.Pointer(&util)))
+	if ret != 0 {
+		return 0, 0, 0, false
+	}
+
+	// 获取显存
+	getMem := lib.NewProc("nvmlDeviceGetMemoryInfo")
+	var mem struct {
+		Total uint64
+		Free  uint64
+		Used  uint64
+	}
+	ret, _, _ = getMem.Call(device, uintptr(unsafe.Pointer(&mem)))
+
+	// 获取功耗 (单位通常是毫瓦)
+	getPower := lib.NewProc("nvmlDeviceGetPowerUsage")
+	var power uint32
+	ret, _, _ = getPower.Call(device, uintptr(unsafe.Pointer(&power)))
+
+	return float64(util.GPU), mem.Used, float64(power) / 1000.0, true
+}

modules/antigravity-api/storage.js

@@ -690,10 +690,24 @@ function getStats() {
       .get();
 
     const accounts = getAccounts();
+
+    // 获取最近 14 天的趋势数据
+    const dailyTrend = db.prepare(`
+        SELECT 
+            strftime('%Y-%m-%d', datetime(created_at, 'localtime')) as date,
+            COUNT(*) as total,
+            SUM(CASE WHEN status_code = 200 THEN 1 ELSE 0 END) as success
+        FROM antigravity_logs
+        WHERE created_at >= datetime('now', '-14 days', 'localtime')
+        GROUP BY date
+        ORDER BY date ASC
+    `).all();
+
     return {
       total_calls: stats.total_calls || 0,
       success_calls: stats.success_calls || 0,
       fail_calls: stats.fail_calls || 0,
+      daily_trend: dailyTrend || [],
       accounts: {
         total: accounts.length,
         online: accounts.filter(a => a.status === 'online').length,