check-smoke: add node name label to metrics (#529)

Author: darkobas2

pkg/check/smoke/metrics.go

@@ -19,7 +19,10 @@ type metrics struct {
 	DownloadThroughput  *prometheus.GaugeVec
 }
 
-const labelSizeBytes = "size_bytes"
+const (
+	labelSizeBytes = "size_bytes"
+	labelNodeName  = "node_name"
+)
 
 func newMetrics(subsystem string) metrics {
 	return metrics{
@@ -46,7 +49,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "upload_attempts",
 				Help:      "Number of upload attempts.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		DownloadAttempts: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
@@ -55,7 +58,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "download_attempts",
 				Help:      "Number of download attempts.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		UploadErrors: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
@@ -64,7 +67,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "upload_errors_count",
 				Help:      "The total number of errors encountered before successful upload.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		DownloadErrors: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
@@ -73,7 +76,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "download_errors_count",
 				Help:      "The total number of errors encountered before successful download.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		DownloadMismatch: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
@@ -82,7 +85,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "download_mismatch",
 				Help:      "The total number of times uploaded data is different from downloaded data.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		UploadDuration: prometheus.NewHistogramVec(
 			prometheus.HistogramOpts{
@@ -91,7 +94,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "data_upload_duration",
 				Help:      "Data upload duration through the /bytes endpoint.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		DownloadDuration: prometheus.NewHistogramVec(
 			prometheus.HistogramOpts{
@@ -100,7 +103,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "data_download_duration",
 				Help:      "Data download duration through the /bytes endpoint.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		UploadThroughput: prometheus.NewGaugeVec(
 			prometheus.GaugeOpts{
@@ -109,7 +112,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "upload_throughput_bytes_per_second",
 				Help:      "Upload throughput in bytes per second.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 		DownloadThroughput: prometheus.NewGaugeVec(
 			prometheus.GaugeOpts{
@@ -118,7 +121,7 @@ func newMetrics(subsystem string) metrics {
 				Name:      "download_throughput_bytes_per_second",
 				Help:      "Download throughput in bytes per second.",
 			},
-			[]string{labelSizeBytes},
+			[]string{labelSizeBytes, labelNodeName},
 		),
 	}
 }

pkg/check/smoke/smoke.go

@@ -189,10 +189,10 @@ func (c *Check) run(ctx context.Context, cluster orchestration.Cluster, o Option
 
 				c.logger.WithField("batch_id", batchID).Infof("using batch for size %d", contentSize)
 
-				c.metrics.UploadAttempts.WithLabelValues(sizeLabel).Inc()
+				c.metrics.UploadAttempts.WithLabelValues(sizeLabel, txName).Inc()
 				address, txDuration, err = test.upload(txCtx, txName, txData, batchID)
 				if err != nil {
-					c.metrics.UploadErrors.WithLabelValues(sizeLabel).Inc()
+					c.metrics.UploadErrors.WithLabelValues(sizeLabel, txName).Inc()
 					c.logger.Infof("upload failed for size %d: %v", contentSize, err)
 					c.logger.Infof("retrying in: %v", o.TxOnErrWait)
 				} else {
@@ -206,12 +206,12 @@ func (c *Check) run(ctx context.Context, cluster orchestration.Cluster, o Option
 				continue
 			}
 
-			c.metrics.UploadDuration.WithLabelValues(sizeLabel).Observe(txDuration.Seconds())
+			c.metrics.UploadDuration.WithLabelValues(sizeLabel, txName).Observe(txDuration.Seconds())
 
 			// Calculate and record upload throughput in bytes per second
 			if txDuration.Seconds() > 0 {
 				uploadThroughput := float64(contentSize) / txDuration.Seconds()
-				c.metrics.UploadThroughput.WithLabelValues(sizeLabel).Set(uploadThroughput)
+				c.metrics.UploadThroughput.WithLabelValues(sizeLabel, txName).Set(uploadThroughput)
 			}
 			c.logger.Infof("upload completed for size %d in %s", contentSize, txDuration)
 
@@ -231,32 +231,32 @@ func (c *Check) run(ctx context.Context, cluster orchestration.Cluster, o Option
 				case <-time.After(o.RxOnErrWait):
 				}
 
-				c.metrics.DownloadAttempts.WithLabelValues(sizeLabel).Inc()
+				c.metrics.DownloadAttempts.WithLabelValues(sizeLabel, rxName).Inc()
 
 				rxCtx, rxCancel = context.WithTimeout(ctx, o.DownloadTimeout)
 				rxData, rxDuration, err = test.download(rxCtx, rxName, address)
 				if err != nil {
-					c.metrics.DownloadErrors.WithLabelValues(sizeLabel).Inc()
+					c.metrics.DownloadErrors.WithLabelValues(sizeLabel, rxName).Inc()
 					c.logger.Infof("download failed for size %d: %v", contentSize, err)
 					c.logger.Infof("retrying in: %v", o.RxOnErrWait)
 					continue
 				}
 
 				// good download
 				if bytes.Equal(rxData, txData) {
-					c.metrics.DownloadDuration.WithLabelValues(sizeLabel).Observe(rxDuration.Seconds())
+					c.metrics.DownloadDuration.WithLabelValues(sizeLabel, rxName).Observe(rxDuration.Seconds())
 
 					if rxDuration.Seconds() > 0 {
 						downloadThroughput := float64(contentSize) / rxDuration.Seconds()
-						c.metrics.DownloadThroughput.WithLabelValues(sizeLabel).Set(downloadThroughput)
+						c.metrics.DownloadThroughput.WithLabelValues(sizeLabel, rxName).Set(downloadThroughput)
 					}
 					c.logger.Infof("download completed for size %d in %s", contentSize, rxDuration)
 					break
 				}
 
 				// bad download
 				c.logger.Infof("uploaded data does not match downloaded data for size %d", contentSize)
-				c.metrics.DownloadMismatch.WithLabelValues(sizeLabel).Inc()
+				c.metrics.DownloadMismatch.WithLabelValues(sizeLabel, rxName).Inc()
 
 				rxLen, txLen := len(rxData), len(txData)
 				if rxLen != txLen {