From a7afc031563330ed18badd55e05f2a5d2d1d2062 Mon Sep 17 00:00:00 2001 From: Luke Date: Fri, 10 May 2024 15:07:12 +0900 Subject: [PATCH] chore: add hostname to spot int metrics --- pkg/controllers/interruption/controller.go | 2 +- pkg/controllers/interruption/metrics.go | 4 +++- 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/pkg/controllers/interruption/controller.go b/pkg/controllers/interruption/controller.go index c08612b2c847..5cd75edaa8bf 100644 --- a/pkg/controllers/interruption/controller.go +++ b/pkg/controllers/interruption/controller.go @@ -198,7 +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() + spotTotal.WithLabelValues(instanceType, zone, nodeClaim.Status.NodeName, nodeClaim.Labels["karpenter.sh/nodepool"]).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 index 5ba2dda465c6..e414dbab5022 100644 --- a/pkg/controllers/interruption/metrics.go +++ b/pkg/controllers/interruption/metrics.go @@ -28,6 +28,8 @@ const ( terminationReasonLabel = "interruption" instanceTypeLabel = "instance_type" zoneLabel = "zone" + hostLabel = "node_name" + poolLabel = "node_pool" ) var ( @@ -73,7 +75,7 @@ var ( Name: "spot_total", Help: "Number of the spot interruption. Labeled by AZ, instance type", }, - []string{instanceTypeLabel, zoneLabel}, + []string{instanceTypeLabel, zoneLabel, hostLabel, poolLabel}, ) )