Merge pull request #8 from smartnews/spot-metrics

chore: add spot interruption metrics with zone and instance type
smartnews · May 8, 2024 · 1f59c19 · 1f59c19
2 parents dd236c8 + 8f90524
commit 1f59c19
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 0 deletions.
diff --git a/pkg/controllers/interruption/controller.go b/pkg/controllers/interruption/controller.go
@@ -198,6 +198,7 @@ func (c *Controller) handleNodeClaim(ctx context.Context, msg messages.Message,
 		if zone != "" && instanceType != "" {
 			c.unavailableOfferingsCache.MarkUnavailable(ctx, string(msg.Kind()), instanceType, zone, v1beta1.CapacityTypeSpot)
 		}
+		spotTotal.WithLabelValues(instanceType, zone).Inc()
 		// try to create a new nodeclaim immediately but ignore error if it fails
 		if err := c.createNodeClaim(ctx, nodeClaim); err != nil {
 			logging.FromContext(ctx).Errorf("[interruption handling]failed to create a new nodeclaim, %v", err)

diff --git a/pkg/controllers/interruption/metrics.go b/pkg/controllers/interruption/metrics.go
@@ -26,6 +26,8 @@ const (
 	messageTypeLabel       = "message_type"
 	actionTypeLabel        = "action_type"
 	terminationReasonLabel = "interruption"
+	instanceTypeLabel      = "instance_type"
+	zoneLabel              = "zone"
 )
 
 var (
@@ -64,6 +66,15 @@ var (
 		},
 		[]string{actionTypeLabel},
 	)
+	spotTotal = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: metrics.Namespace,
+			Subsystem: interruptionSubsystem,
+			Name:      "spot_total",
+			Help:      "Number of the spot interruption. Labeled by AZ, instance type",
+		},
+		[]string{instanceTypeLabel, zoneLabel},
+	)
 )
 
 func init() {