fix: create batch failures should not count toward upload error

ethersphere · Oct 2, 2023 · c7273bf · c7273bf
1 parent 8017852
commit c7273bf
Show file tree

Hide file tree

Showing 2 changed files with 31 additions and 12 deletions.
diff --git a/pkg/check/smoke/metrics.go b/pkg/check/smoke/metrics.go
@@ -6,17 +6,35 @@ import (
 )
 
 type metrics struct {
-	UploadErrors     prometheus.Counter
-	UploadAttempts   prometheus.Counter
-	DownloadErrors   prometheus.Counter
-	DownloadMismatch prometheus.Counter
-	DownloadAttempts prometheus.Counter
-	UploadDuration   prometheus.Histogram
-	DownloadDuration prometheus.Histogram
+	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
 }
 
 func newMetrics(subsystem string) metrics {
 	return metrics{
+		BatchCreateAttempts: prometheus.NewCounter(
+			prometheus.CounterOpts{
+				Namespace: m.Namespace,
+				Subsystem: subsystem,
+				Name:      "batch_create_attemps",
+				Help:      "Number of batch create attemps.",
+			},
+		),
+		BatchCreateErrors: prometheus.NewCounter(
+			prometheus.CounterOpts{
+				Namespace: m.Namespace,
+				Subsystem: subsystem,
+				Name:      "batch_create_errors",
+				Help:      "Total errors encountered while creating batches.",
+			},
+		),
 		UploadAttempts: prometheus.NewCounter(
 			prometheus.CounterOpts{
 				Namespace: m.Namespace,

diff --git a/pkg/check/smoke/smoke.go b/pkg/check/smoke/smoke.go
@@ -146,20 +146,21 @@ func (c *Check) Run(ctx context.Context, cluster orchestration.Cluster, opts int
 			case <-time.After(o.TxOnErrWait):
 			}
 
-			c.metrics.UploadAttempts.Inc()
-
-			txCtx, txCancel = context.WithTimeout(ctx, o.UploadTimeout)
-
+			c.metrics.BatchCreateAttempts.Inc()
 			batchID := batches.Get(txName)
 			if batchID == "" {
-				batchID, err = clients[txName].CreatePostageBatch(txCtx, o.PostageAmount, o.PostageDepth, "load-test", true)
+				batchID, err = clients[txName].CreatePostageBatch(context.Background(), o.PostageAmount, o.PostageDepth, "load-test", true)
 				if err != nil {
 					c.logger.Errorf("create new batch: %v", err)
+					c.metrics.BatchCreateErrors.Inc()
 					continue
 				}
 				batches.Store(txName, batchID)
 			}
 
+			txCtx, txCancel = context.WithTimeout(ctx, o.UploadTimeout)
+
+			c.metrics.UploadAttempts.Inc()
 			address, txDuration, err = test.upload(txCtx, txName, txData, batchID)
 			if err != nil {
 				c.metrics.UploadErrors.Inc()