fix job active count; count terminating pod as failed (#214)

Signed-off-by: yowenter <[email protected]>

Author: yowenter

pkg/controller.v1/common/status_test.go

@@ -2,10 +2,12 @@ package common
 
 import (
 	"testing"
+	"time"
 
 	apiv1 "github.com/kubeflow/common/pkg/apis/common/v1"
 	"github.com/stretchr/testify/assert"
 	corev1 "k8s.io/api/core/v1"
+	metaV1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
 func TestUpdateJobReplicaStatuses(t *testing.T) {
@@ -14,13 +16,14 @@ func TestUpdateJobReplicaStatuses(t *testing.T) {
 	_, ok := jobStatus.ReplicaStatuses["worker"]
 	// assert ReplicaStatus for "worker" exists
 	assert.True(t, ok)
-	setStatusForTest(&jobStatus, "worker", 2, 3, 1)
-	assert.Equal(t, jobStatus.ReplicaStatuses["worker"].Failed, int32(2))
+	setStatusForTest(&jobStatus, "worker", 2, 3, 1, 1)
+	// terminating pod should count as failed.
+	assert.Equal(t, jobStatus.ReplicaStatuses["worker"].Failed, int32(3))
 	assert.Equal(t, jobStatus.ReplicaStatuses["worker"].Succeeded, int32(3))
 	assert.Equal(t, jobStatus.ReplicaStatuses["worker"].Active, int32(1))
 }
 
-func setStatusForTest(jobStatus *apiv1.JobStatus, rtype apiv1.ReplicaType, failed, succeeded, active int32) {
+func setStatusForTest(jobStatus *apiv1.JobStatus, rtype apiv1.ReplicaType, failed, succeeded, active, terminating int32) {
 	pod := corev1.Pod{
 		Status: corev1.PodStatus{},
 	}
@@ -37,4 +40,10 @@ func setStatusForTest(jobStatus *apiv1.JobStatus, rtype apiv1.ReplicaType, faile
 		pod.Status.Phase = corev1.PodRunning
 		updateJobReplicaStatuses(jobStatus, rtype, &pod)
 	}
+	for i = 0; i < terminating; i++ {
+		pod.Status.Phase = corev1.PodRunning
+		deletionTimestamp := metaV1.NewTime(time.Now())
+		pod.DeletionTimestamp = &deletionTimestamp
+		updateJobReplicaStatuses(jobStatus, rtype, &pod)
+	}
 }

pkg/core/status.go

@@ -34,7 +34,14 @@ func InitializeReplicaStatuses(jobStatus *apiv1.JobStatus, rtype apiv1.ReplicaTy
 func UpdateJobReplicaStatuses(jobStatus *apiv1.JobStatus, rtype apiv1.ReplicaType, pod *corev1.Pod) {
 	switch pod.Status.Phase {
 	case corev1.PodRunning:
-		jobStatus.ReplicaStatuses[rtype].Active++
+		if pod.DeletionTimestamp != nil {
+			// when node is not ready, the pod will be in terminating state.
+			// Count deleted Pods as failures to account for orphan Pods that
+			// never have a chance to reach the Failed phase.
+			jobStatus.ReplicaStatuses[rtype].Failed++
+		} else {
+			jobStatus.ReplicaStatuses[rtype].Active++
+		}
 	case corev1.PodSucceeded:
 		jobStatus.ReplicaStatuses[rtype].Succeeded++
 	case corev1.PodFailed: