Merge pull request #2542 from Indresh2410/add-kube-job-status-suspend…

…ed-metric feat: Add new metric kube_job_status_suspended
kubernetes · Nov 7, 2024 · 5e7d0f4 · 5e7d0f4
2 parents c87b507 + 7a54a8e
commit 5e7d0f4
Show file tree

Hide file tree

Showing 3 changed files with 96 additions and 1 deletion.
diff --git a/docs/metrics/workload/job-metrics.md b/docs/metrics/workload/job-metrics.md
@@ -17,3 +17,4 @@
 | kube_job_complete                     | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt; <br> `condition`=&lt;true\|false\|unknown&gt;                                                                                            | STABLE       |
 | kube_job_failed                       | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt; <br> `condition`=&lt;true\|false\|unknown&gt;                                                                                            | STABLE       |
 | kube_job_created                      | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt;                                                                                                                                          | STABLE       |
+| kube_job_status_suspended     | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt;                                                                                                                                          | EXPERIMENTAL       |
diff --git a/internal/store/job.go b/internal/store/job.go
@@ -26,6 +26,7 @@ import (
 	generator "k8s.io/kube-state-metrics/v2/pkg/metric_generator"
 
 	v1batch "k8s.io/api/batch/v1"
+	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/watch"
@@ -355,6 +356,27 @@ func jobMetricFamilies(allowAnnotationsList, allowLabelsList []string) []generat
 				}
 			}),
 		),
+		*generator.NewFamilyGeneratorWithStability(
+			"kube_job_status_suspended",
+			"The number of pods which reached Phase Suspended.",
+			metric.Gauge,
+			basemetrics.ALPHA,
+			"",
+			wrapJobFunc(func(j *v1batch.Job) *metric.Family {
+				ms := []*metric.Metric{}
+				for _, c := range j.Status.Conditions {
+					if c.Type == v1batch.JobSuspended {
+						ms = append(ms, &metric.Metric{
+							Value: boolFloat64(c.Status == v1.ConditionTrue),
+						})
+					}
+				}
+
+				return &metric.Family{
+					Metrics: ms,
+				}
+			}),
+		),
 		*generator.NewFamilyGeneratorWithStability(
 			"kube_job_owner",
 			"Information about the Job's owner.",

diff --git a/internal/store/job_test.go b/internal/store/job_test.go
@@ -44,6 +44,7 @@ var (
 
 func TestJobStore(t *testing.T) {
 	var trueValue = true
+	var falseValue = false
 
 	// Fixed metadata on type and help text. We prepend this to every expected
 	// output so we only have to modify a single place when doing adjustments.
@@ -77,7 +78,10 @@ func TestJobStore(t *testing.T) {
 		# HELP kube_job_status_start_time [STABLE] StartTime represents time when the job was acknowledged by the Job Manager.
 		# TYPE kube_job_status_start_time gauge
 		# HELP kube_job_status_succeeded [STABLE] The number of pods which reached Phase Succeeded.
-		# TYPE kube_job_status_succeeded gauge`
+		# TYPE kube_job_status_succeeded gauge
+		# HELP kube_job_status_suspended The number of pods which reached Phase Suspended.
+		# TYPE kube_job_status_suspended gauge
+		`
 
 	cases := []generateMetricsTestCase{
 		{
@@ -272,6 +276,74 @@ func TestJobStore(t *testing.T) {
 				kube_job_status_failed{job_name="SuccessfulJob2NoActiveDeadlineSeconds",namespace="ns1"} 0
 				kube_job_status_start_time{job_name="SuccessfulJob2NoActiveDeadlineSeconds",namespace="ns1"} 1.495800607e+09
 				kube_job_status_succeeded{job_name="SuccessfulJob2NoActiveDeadlineSeconds",namespace="ns1"} 1
+`,
+		},
+		{
+			Obj: &v1batch.Job{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "SuspendedNoActiveDeadlineSeconds",
+					Namespace:  "ns1",
+					Generation: 1,
+				},
+				Status: v1batch.JobStatus{
+					Active:    0,
+					Failed:    0,
+					Succeeded: 0,
+					StartTime: &metav1.Time{Time: SuccessfulJob2StartTime},
+					Conditions: []v1batch.JobCondition{
+						{Type: v1batch.JobSuspended, Status: v1.ConditionTrue},
+					},
+				},
+				Spec: v1batch.JobSpec{
+					Suspend:     &trueValue,
+					Parallelism: &Parallelism1,
+					Completions: &Completions1,
+				},
+			},
+			Want: metadata + `
+				kube_job_owner{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1",owner_is_controller="",owner_kind="",owner_name=""} 1
+				kube_job_info{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_spec_completions{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_spec_parallelism{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_status_active{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_failed{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_start_time{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1.495800607e+09
+				kube_job_status_succeeded{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+                kube_job_status_suspended{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+`,
+		},
+		{
+			Obj: &v1batch.Job{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "UnsuspendedNoActiveDeadlineSeconds",
+					Namespace:  "ns1",
+					Generation: 1,
+				},
+				Status: v1batch.JobStatus{
+					Active:    0,
+					Failed:    0,
+					Succeeded: 0,
+					StartTime: &metav1.Time{Time: SuccessfulJob2StartTime},
+					Conditions: []v1batch.JobCondition{
+						{Type: v1batch.JobSuspended, Status: v1.ConditionFalse},
+					},
+				},
+				Spec: v1batch.JobSpec{
+					Suspend:     &falseValue,
+					Parallelism: &Parallelism1,
+					Completions: &Completions1,
+				},
+			},
+			Want: metadata + `
+				kube_job_owner{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1",owner_is_controller="",owner_kind="",owner_name=""} 1
+				kube_job_info{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_spec_completions{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_spec_parallelism{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_status_active{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_failed{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_start_time{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1.495800607e+09
+				kube_job_status_succeeded{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+                kube_job_status_suspended{job_name="UnsuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
 `,
 		},
 	}