feat: add CreateJobCount and CreateJobSuccessCount metrics (#3588)

Signed-off-by: Gaius <gaius.qi@gmail.com>
dragonflyoss · Oct 18, 2024 · 2793851 · 2793851
1 parent 030f337
commit 2793851
Show file tree

Hide file tree

Showing 3 changed files with 21 additions and 0 deletions.
diff --git a/manager/handlers/job.go b/manager/handlers/job.go
@@ -23,6 +23,7 @@ import (
 	"github.com/gin-gonic/gin/binding"
 
 	"d7y.io/dragonfly/v2/internal/job"
+	"d7y.io/dragonfly/v2/manager/metrics"
 	_ "d7y.io/dragonfly/v2/manager/models" // nolint
 	"d7y.io/dragonfly/v2/manager/types"
 )
@@ -45,6 +46,8 @@ func (h *Handlers) CreateJob(ctx *gin.Context) {
 		return
 	}
 
+	// Collect CreateJobCount metrics.
+	metrics.CreateJobCount.WithLabelValues(json.Type).Inc()
 	switch json.Type {
 	case job.PreheatJob:
 		var json types.CreatePreheatJobRequest

diff --git a/manager/metrics/metrics.go b/manager/metrics/metrics.go
@@ -46,6 +46,20 @@ var (
 		Help:      "Counter of the number of failed of searching scheduler cluster.",
 	}, []string{"version", "commit"})
 
+	CreateJobCount = promauto.NewCounterVec(prometheus.CounterOpts{
+		Namespace: types.MetricsNamespace,
+		Subsystem: types.ManagerMetricsName,
+		Name:      "create_job_total",
+		Help:      "Counter of the number of creating job.",
+	}, []string{"name"})
+
+	CreateJobSuccessCount = promauto.NewCounterVec(prometheus.CounterOpts{
+		Namespace: types.MetricsNamespace,
+		Subsystem: types.ManagerMetricsName,
+		Name:      "create_job_success_total",
+		Help:      "Counter of the number of succeeded of creating job.",
+	}, []string{"name"})
+
 	VersionGauge = promauto.NewGaugeVec(prometheus.GaugeOpts{
 		Namespace: types.MetricsNamespace,
 		Subsystem: types.ManagerMetricsName,

diff --git a/manager/service/job.go b/manager/service/job.go
@@ -25,6 +25,7 @@ import (
 
 	logger "d7y.io/dragonfly/v2/internal/dflog"
 	internaljob "d7y.io/dragonfly/v2/internal/job"
+	"d7y.io/dragonfly/v2/manager/metrics"
 	"d7y.io/dragonfly/v2/manager/models"
 	"d7y.io/dragonfly/v2/manager/types"
 	"d7y.io/dragonfly/v2/pkg/retry"
@@ -311,6 +312,9 @@ func (s *service) pollingJob(ctx context.Context, name string, id uint, groupID
 
 		switch job.State {
 		case machineryv1tasks.StateSuccess:
+			// Collect CreateJobSuccessCount. metrics.
+			metrics.CreateJobSuccessCount.WithLabelValues(name).Inc()
+
 			log.Info("polling group succeeded")
 			return nil, true, nil
 		case machineryv1tasks.StateFailure: