fix: account for queued loadgen rate starts

README.md

@@ -263,7 +263,8 @@ go run ./cmd/proxy-loadgen `
 连接握手延迟分位上界、吞吐和 Go 运行时内存/GC 快照。它不会输出提取响应中的地址或
 凭据，也不构成 100,000 QPS 证明。限速场景以 `Requests` 表示实际发起数；当
 `RateStartsDropped` 非零时，目标速率受压测端并发容量或时间窗限制，报告吞吐不得
-标注为已达到配置的 `-rate`。
+标注为已达到配置的 `-rate`。每个生成的限速令牌均会计入实际发起或丢弃，因此
+`RateStartsGenerated = Requests + RateStartsDropped`。
 需要保留单次容量证据时，使用 `-output REPORT_FILE` 同时写出 JSON 文件；文件在
 同目录完整写入后才替换目标，标准输出仍保留相同报告，便于交给日志或指标系统。
 

deploy/kubernetes/base/configmap.yaml

@@ -21,6 +21,16 @@ data:
       limits:
         maxConcurrentConnections: 100000
         requestsPerMinutePerClient: 60000
+      transport:
+        dialTimeout: 10s
+        handshakeTimeout: 15s
+        responseHeaderTimeout: 30s
+        idleConnTimeout: 90s
+        maxIdleConns: 20000
+        maxIdleConnsPerHost: 32
+        maxConnsPerHost: 32
+        tunnelBufferBytes: 32768
+        tunnelIdleTimeout: 5m
       retry:
         maxAttempts: 2
         retryMethods: [GET, HEAD]

internal/config/config_test.go

@@ -204,6 +204,7 @@ func TestShippedDeploymentConfigurationsResolveEnvironment(t *testing.T) {
 		wantGatewayUser string
 		wantPostgresURL string
 		wantRedisURL    string
+		wantTransport   GatewayTransport
 	}{
 		{
 			name:            "local",
@@ -211,6 +212,12 @@ func TestShippedDeploymentConfigurationsResolveEnvironment(t *testing.T) {
 			wantGatewayUser: "local-gateway",
 			wantPostgresURL: "postgres://proxy_pool:local-only-change-me@postgres:5432/proxy_pool?sslmode=disable",
 			wantRedisURL:    "redis://redis:6379/0",
+			wantTransport: GatewayTransport{
+				MaxIdleConns:        20000,
+				MaxIdleConnsPerHost: 32,
+				MaxConnsPerHost:     32,
+				TunnelBufferBytes:   32768,
+			},
 		},
 		{
 			name:            "kubernetes",
@@ -219,6 +226,17 @@ func TestShippedDeploymentConfigurationsResolveEnvironment(t *testing.T) {
 			wantGatewayUser: "resolved-gateway-user",
 			wantPostgresURL: "postgres://resolved",
 			wantRedisURL:    "redis://resolved",
+			wantTransport: GatewayTransport{
+				DialTimeout:           Duration(10 * time.Second),
+				HandshakeTimeout:      Duration(15 * time.Second),
+				ResponseHeaderTimeout: Duration(30 * time.Second),
+				IdleConnTimeout:       Duration(90 * time.Second),
+				MaxIdleConns:          20000,
+				MaxIdleConnsPerHost:   32,
+				MaxConnsPerHost:       32,
+				TunnelBufferBytes:     32768,
+				TunnelIdleTimeout:     Duration(5 * time.Minute),
+			},
 		},
 	}
 	for _, test := range tests {
@@ -248,6 +266,7 @@ func TestShippedDeploymentConfigurationsResolveEnvironment(t *testing.T) {
 				cfg.Admin.Auth.Token != "resolved-admin-token" ||
 				cfg.Storage.PostgresURL != test.wantPostgresURL ||
 				cfg.Storage.RedisURL != test.wantRedisURL ||
+				cfg.Gateway.Transport != test.wantTransport ||
 				cfg.Upstreams["provider-a"].API.Auth.Value != "resolved-provider-a-token" ||
 				cfg.Upstreams["provider-b"].API.Auth.Value != "resolved-provider-b-token" {
 				t.Fatalf("deployment values were not resolved: %+v", cfg.Redacted())

internal/loadgen/http.go

@@ -331,7 +331,7 @@ func runForDuration(ctx context.Context, options Options, execute func(context.C
 	for {
 		select {
 		case <-workloadContext.Done():
-			workers.Wait()
+			waitForRateWorkers(&workers, jobs, stats)
 			return
 		case now := <-ticker.C:
 			credit += float64(options.Rate) * now.Sub(last).Seconds()
@@ -352,7 +352,7 @@ func runForDuration(ctx context.Context, options Options, execute func(context.C
 				case jobs <- struct{}{}:
 				case <-workloadContext.Done():
 					stats.rateStartsDropped.Add(uint64(tokens - index))
-					workers.Wait()
+					waitForRateWorkers(&workers, jobs, stats)
 					return
 				}
 			}
@@ -360,6 +360,11 @@ func runForDuration(ctx context.Context, options Options, execute func(context.C
 	}
 }
 
+func waitForRateWorkers(workers *sync.WaitGroup, jobs <-chan struct{}, stats *counters) {
+	workers.Wait()
+	stats.rateStartsDropped.Add(uint64(len(jobs)))
+}
+
 type requestBuilder func(context.Context) (*http.Request, responseValidator, error)
 
 type responseValidator func(*http.Response) error

internal/loadgen/http_test.go

@@ -66,11 +66,26 @@ func TestRunTimeBoxedRateIsBounded(t *testing.T) {
 		TargetURL: server.URL, Duration: 100 * time.Millisecond, Rate: 100, Concurrency: 2, RequestTimeout: time.Second,
 	})
 	if err != nil || report.Requests == 0 || report.Requests > 20 || report.Requests != report.Completed ||
-		report.Succeeded != report.Completed || requests.Load() != int64(report.Requests) {
+		report.Succeeded+report.Failed != report.Completed || report.RateStartsGenerated != report.Requests+report.RateStartsDropped ||
+		requests.Load() > int64(report.Requests) {
 		t.Fatalf("Run() = (%+v, %v); handler requests=%d", report, err, requests.Load())
 	}
 }
 
+func TestRunForDurationAccountsQueuedRateStartsAtDeadline(t *testing.T) {
+	stats := &counters{}
+	runForDuration(context.Background(), Options{
+		Duration: 40 * time.Millisecond, Rate: 1000, Concurrency: 1,
+	}, func(ctx context.Context) {
+		stats.requests.Add(1)
+		<-ctx.Done()
+	}, stats)
+
+	if generated, accounted := stats.rateStartsGenerated.Load(), stats.requests.Load()+stats.rateStartsDropped.Load(); generated == 0 || generated != accounted {
+		t.Fatalf("rate starts generated=%d, accounted=%d", generated, accounted)
+	}
+}
+
 func TestRunReportsRateStartsDroppedByConcurrency(t *testing.T) {
 	started := make(chan struct{}, 1)
 	server := httptest.NewServer(http.HandlerFunc(func(response http.ResponseWriter, request *http.Request) {