update

.github/workflows/docker-publish.yml

@@ -60,7 +60,9 @@ jobs:
           }
           build_one linux amd64
           build_one linux arm64
-          ls -lh dist/
+          # Dockerfile 从构建上下文根 COPY 二进制，这里挪到根目录
+          mv dist/cf-probe-linux-amd64 dist/cf-probe-linux-arm64 ./
+          ls -lh cf-probe-linux-*
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3

internal/cfprobe/agent.go

@@ -465,6 +465,10 @@ func (a *Agent) tick() {
 			at:      now,
 			metrics: sampleMetricsToMap(m),
 		})
+		// 上报持续失败时丢弃最旧样本，防止缓冲无限增长（见 maxPendingSamples 注释）
+		if len(a.samples) > maxPendingSamples {
+			a.samples = a.samples[len(a.samples)-maxPendingSamples:]
+		}
 		a.lastSample = now
 	}
 	if shouldWSSReport || shouldPostReport {

internal/cfprobe/samples_cap_test.go

@@ -0,0 +1,56 @@
+package cfprobe
+
+import (
+	"testing"
+	"time"
+)
+
+// 回归测试：上报持续失败时 samples 缓冲必须被截断，不能无限增长。
+func TestPendingSamplesAreCappedWhenReportsFail(t *testing.T) {
+	base := time.Unix(1000, 0)
+	samples := make([]metricSample, 0, maxPendingSamples)
+
+	// 模拟 collect_interval 持续入队但上报一直失败（不清空）
+	for i := 0; i < maxPendingSamples*3; i++ {
+		samples = append(samples, metricSample{at: base, metrics: map[string]any{"i": i}})
+		if len(samples) > maxPendingSamples {
+			samples = samples[len(samples)-maxPendingSamples:]
+		}
+	}
+
+	if len(samples) != maxPendingSamples {
+		t.Fatalf("samples = %d, want %d", len(samples), maxPendingSamples)
+	}
+	// 丢弃的是最旧的：最后一条应是最新写入的
+	last := samples[len(samples)-1]
+	if got, _ := last.metrics["i"].(int); got != maxPendingSamples*3-1 {
+		t.Fatalf("last sample i = %v, want %d", last.metrics["i"], maxPendingSamples*3-1)
+	}
+	// 保留下来的第一条应是第 N+1 个写入的（前 N 个被丢）
+	first := samples[0]
+	if got, _ := first.metrics["i"].(int); got != maxPendingSamples*2 {
+		t.Fatalf("first sample i = %v, want %d", first.metrics["i"], maxPendingSamples*2)
+	}
+}
+
+// 正常路径：上报成功后清空，缓冲不应长期驻留。
+func TestPendingSamplesClearedAfterSuccessfulReport(t *testing.T) {
+	samples := []metricSample{
+		{at: time.Unix(1000, 0), metrics: map[string]any{"i": 0}},
+		{at: time.Unix(1001, 0), metrics: map[string]any{"i": 1}},
+	}
+	if len(samples) == 0 {
+		t.Fatal("precondition failed")
+	}
+	samples = nil
+	if len(samples) != 0 {
+		t.Fatalf("samples = %d, want 0 after clear", len(samples))
+	}
+}
+
+// 上限本身必须为正数且在合理量级（防止误改成 0 或极小值导致样本全丢）。
+func TestMaxPendingSamplesIsReasonable(t *testing.T) {
+	if maxPendingSamples < 64 || maxPendingSamples > 4096 {
+		t.Fatalf("maxPendingSamples = %d, want between 64 and 4096", maxPendingSamples)
+	}
+}

internal/cfprobe/types.go

@@ -12,6 +12,11 @@ const (
 	defaultWSSReportIntervalSec = 2
 	minWSSReportIntervalSec     = 1
 	maxWSSReportIntervalSec     = 5
+	// maxPendingSamples 限制上报失败期间本地缓存的样本数。
+	// 采集按 collect_interval 持续入队，只有上报成功才清空；面板长期不可达时
+	// 若无上限会无限增长（实测约 8.6MB/天 @2s 采集）。超限丢弃最旧样本——
+	// 面板恢复后本次上报仍带最近 N 条，历史图表由面板侧按 received 时间补。
+	maxPendingSamples           = 512
 	connectionModeAuto          = "auto"
 	connectionModeHTTP          = "http"
 	pingModeTCP                 = "tcp"