feat: smoke file sizes (#505)

* feat: fize-sizes param for smoke

* ci: file-sizes in configs

* ci: file-sizes in staging

* refactor: comments

* refactor: backward compatibility for ContentSize

* fix: comments

* fix(load): use content-size for load check (#522)

---------

Co-authored-by: Ljubiša Gačević <35105035+gacevicljubisa@users.noreply.github.com>

Author: nugaon

config/config.yaml

@@ -335,7 +335,7 @@ checks:
     type: settlements
   smoke:
     options:
-      content-size: 5000
+      file-sizes: [31457280, 104857600, 314572800] # [30MB, 100MB, 300MB]  
       postage-ttl: 72h
       postage-depth: 21
       postage-label: test-label
@@ -369,7 +369,7 @@ checks:
     type: withdraw
   load:
     options:
-      content-size: 500000 #5000000
+      content-size: 500000
       gas-price: 10000000000
       postage-ttl: 72h
       postage-depth: 21

config/local.yaml

@@ -282,7 +282,8 @@ checks:
     type: settlements
   ci-smoke:
     options:
-      content-size: 50000
+      content-size: 3072
+      file-sizes: [3072, 30720, 307200, 3145728] # [3KB, 30KB, 300KB, 3MB]
       postage-ttl: 24h
       postage-depth: 21
       postage-label: test-label
@@ -295,7 +296,7 @@ checks:
     type: smoke
   ci-load:
     options:
-      content-size: 50000
+      content-size: 3072
       postage-ttl: 24h
       postage-depth: 21
       postage-label: test-label

config/public-testnet.yaml

@@ -176,7 +176,7 @@ checks:
     type: load
   pt-smoke:
     options:
-      content-size: 50000000
+      file-sizes: [3072, 30720, 307200, 31457280, 104857600, 314572800] # [3KB, 30KB, 300KB, 30MB, 100MB, 300MB]
       postage-ttl: 360h # 15 days
       postage-depth: 22
       postage-label: test-label

pkg/check/smoke/load.go

@@ -60,8 +60,10 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 		return errors.New("max committed depth is not set")
 	}
 
+	contentSize := o.ContentSize
+
 	c.logger.Infof("random seed: %v", o.RndSeed)
-	c.logger.Infof("content size: %v", o.ContentSize)
+	c.logger.Infof("content size: %v", contentSize)
 	c.logger.Infof("max committed depth: %v", o.MaxCommittedDepth)
 	c.logger.Infof("committed depth check wait time: %v", o.CommittedDepthCheckWait)
 	c.logger.Infof("total duration: %s", o.Duration.String())
@@ -82,18 +84,20 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 			c.logger.Info("we are done")
 			return nil
 		default:
-			c.logger.Infof("starting iteration: #%d", i)
+			c.logger.Infof("starting iteration: #%d bytes (%.2f KB)", contentSize, float64(contentSize)/1024)
 		}
 
+		sizeLabel := fmt.Sprintf("%d", contentSize)
+
 		var (
 			txDuration time.Duration
 			txData     []byte
 			address    swarm.Address
 		)
 
-		txData = make([]byte, o.ContentSize)
+		txData = make([]byte, contentSize)
 		if _, err := crand.Read(txData); err != nil {
-			c.logger.Infof("unable to create random content: %v", err)
+			c.logger.Infof("unable to create random content for size %d: %v", contentSize, err)
 			continue
 		}
 
@@ -125,7 +129,7 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 						return
 					}
 
-					c.metrics.UploadAttempts.Inc()
+					c.metrics.UploadAttempts.WithLabelValues(sizeLabel).Inc()
 					var duration time.Duration
 					c.logger.Infof("uploading to: %s", txName)
 
@@ -139,7 +143,7 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 
 					address, duration, err = test.upload(ctx, txName, txData, batchID)
 					if err != nil {
-						c.metrics.UploadErrors.Inc()
+						c.metrics.UploadErrors.WithLabelValues(sizeLabel).Inc()
 						c.logger.Infof("upload failed: %v", err)
 						c.logger.Infof("retrying in: %v", o.TxOnErrWait)
 						time.Sleep(o.TxOnErrWait)
@@ -183,11 +187,11 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 					default:
 					}
 
-					c.metrics.DownloadAttempts.Inc()
+					c.metrics.DownloadAttempts.WithLabelValues(sizeLabel).Inc()
 
 					rxData, rxDuration, err = test.download(ctx, rxName, address)
 					if err != nil {
-						c.metrics.DownloadErrors.Inc()
+						c.metrics.DownloadErrors.WithLabelValues(sizeLabel).Inc()
 						c.logger.Infof("download failed: %v", err)
 						c.logger.Infof("retrying in: %v", o.RxOnErrWait)
 						time.Sleep(o.RxOnErrWait)
@@ -202,7 +206,7 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 				if !bytes.Equal(rxData, txData) {
 					c.logger.Info("uploaded data does not match downloaded data")
 
-					c.metrics.DownloadMismatch.Inc()
+					c.metrics.DownloadMismatch.WithLabelValues(sizeLabel).Inc()
 
 					rxLen, txLen := len(rxData), len(txData)
 					if rxLen != txLen {
@@ -225,8 +229,16 @@ func (c *LoadCheck) run(ctx context.Context, cluster orchestration.Cluster, o Op
 
 				// We want to update the metrics when no error has been
 				// encountered in order to avoid counter mismatch.
-				c.metrics.UploadDuration.Observe(txDuration.Seconds())
-				c.metrics.DownloadDuration.Observe(rxDuration.Seconds())
+				c.metrics.UploadDuration.WithLabelValues(sizeLabel).Observe(txDuration.Seconds())
+				c.metrics.DownloadDuration.WithLabelValues(sizeLabel).Observe(rxDuration.Seconds())
+				if txDuration.Seconds() > 0 {
+					uploadThroughput := float64(contentSize) / txDuration.Seconds()
+					c.metrics.UploadThroughput.WithLabelValues(sizeLabel).Set(uploadThroughput)
+				}
+				if rxDuration.Seconds() > 0 {
+					downloadThroughput := float64(contentSize) / rxDuration.Seconds()
+					c.metrics.DownloadThroughput.WithLabelValues(sizeLabel).Set(downloadThroughput)
+				}
 			}()
 		}
 

pkg/check/smoke/metrics.go

@@ -8,16 +8,19 @@ import (
 type metrics struct {
 	BatchCreateErrors   prometheus.Counter
 	BatchCreateAttempts prometheus.Counter
-	UploadErrors        prometheus.Counter
-	UploadAttempts      prometheus.Counter
-	DownloadErrors      prometheus.Counter
-	DownloadMismatch    prometheus.Counter
-	DownloadAttempts    prometheus.Counter
-	UploadDuration      prometheus.Histogram
-	DownloadDuration    prometheus.Histogram
-	UploadSize          prometheus.Gauge
+	UploadErrors        *prometheus.CounterVec
+	UploadAttempts      *prometheus.CounterVec
+	DownloadErrors      *prometheus.CounterVec
+	DownloadMismatch    *prometheus.CounterVec
+	DownloadAttempts    *prometheus.CounterVec
+	UploadDuration      *prometheus.HistogramVec
+	DownloadDuration    *prometheus.HistogramVec
+	UploadThroughput    *prometheus.GaugeVec
+	DownloadThroughput  *prometheus.GaugeVec
 }
 
+const labelSizeBytes = "size_bytes"
+
 func newMetrics(subsystem string) metrics {
 	return metrics{
 		BatchCreateAttempts: prometheus.NewCounter(
@@ -36,69 +39,86 @@ func newMetrics(subsystem string) metrics {
 				Help:      "Total errors encountered while creating batches.",
 			},
 		),
-		UploadAttempts: prometheus.NewCounter(
+		UploadAttempts: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "upload_attempts",
 				Help:      "Number of upload attempts.",
 			},
+			[]string{labelSizeBytes},
 		),
-		DownloadAttempts: prometheus.NewCounter(
+		DownloadAttempts: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "download_attempts",
 				Help:      "Number of download attempts.",
 			},
+			[]string{labelSizeBytes},
 		),
-		UploadErrors: prometheus.NewCounter(
+		UploadErrors: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "upload_errors_count",
 				Help:      "The total number of errors encountered before successful upload.",
 			},
+			[]string{labelSizeBytes},
 		),
-		DownloadErrors: prometheus.NewCounter(
+		DownloadErrors: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "download_errors_count",
 				Help:      "The total number of errors encountered before successful download.",
 			},
+			[]string{labelSizeBytes},
 		),
-		DownloadMismatch: prometheus.NewCounter(
+		DownloadMismatch: prometheus.NewCounterVec(
 			prometheus.CounterOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "download_mismatch",
 				Help:      "The total number of times uploaded data is different from downloaded data.",
 			},
+			[]string{labelSizeBytes},
 		),
-		UploadDuration: prometheus.NewHistogram(
+		UploadDuration: prometheus.NewHistogramVec(
 			prometheus.HistogramOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "data_upload_duration",
 				Help:      "Data upload duration through the /bytes endpoint.",
 			},
+			[]string{labelSizeBytes},
 		),
-		DownloadDuration: prometheus.NewHistogram(
+		DownloadDuration: prometheus.NewHistogramVec(
 			prometheus.HistogramOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
 				Name:      "data_download_duration",
 				Help:      "Data download duration through the /bytes endpoint.",
 			},
+			[]string{labelSizeBytes},
+		),
+		UploadThroughput: prometheus.NewGaugeVec(
+			prometheus.GaugeOpts{
+				Namespace: m.Namespace,
+				Subsystem: subsystem,
+				Name:      "upload_throughput_bytes_per_second",
+				Help:      "Upload throughput in bytes per second.",
+			},
+			[]string{labelSizeBytes},
 		),
-		UploadSize: prometheus.NewGauge(
+		DownloadThroughput: prometheus.NewGaugeVec(
 			prometheus.GaugeOpts{
 				Namespace: m.Namespace,
 				Subsystem: subsystem,
-				Name:      "upload_size",
-				Help:      "Amount of data uploaded per upload.",
+				Name:      "download_throughput_bytes_per_second",
+				Help:      "Download throughput in bytes per second.",
 			},
+			[]string{labelSizeBytes},
 		),
 	}
 }