Merge pull request volcano-sh#176 from lminzhw/retain_pod

retain pod with special phase
kevin-wangzefeng · May 16, 2019 · 85e3350 · 85e3350
2 parents 51ee86c + a72a103
commit 85e3350
Show file tree

Hide file tree

Showing 11 changed files with 51 additions and 34 deletions.
diff --git a/pkg/controllers/job/job_controller_actions.go b/pkg/controllers/job/job_controller_actions.go
@@ -35,7 +35,7 @@ import (
 	"volcano.sh/volcano/pkg/controllers/job/state"
 )
 
-func (cc *Controller) killJob(jobInfo *apis.JobInfo, updateStatus state.UpdateStatusFn) error {
+func (cc *Controller) killJob(jobInfo *apis.JobInfo, podRetainPhase state.PhaseMap, updateStatus state.UpdateStatusFn) error {
 	glog.V(3).Infof("Killing Job <%s/%s>", jobInfo.Job.Namespace, jobInfo.Job.Name)
 	defer glog.V(3).Infof("Finished Job <%s/%s> killing", jobInfo.Job.Namespace, jobInfo.Job.Name)
 
@@ -62,20 +62,27 @@ func (cc *Controller) killJob(jobInfo *apis.JobInfo, updateStatus state.UpdateSt
 				continue
 			}
 
-			if err := cc.deleteJobPod(job.Name, pod); err == nil {
-				terminating++
-			} else {
-				errs = append(errs, err)
-				switch pod.Status.Phase {
-				case v1.PodRunning:
-					running++
-				case v1.PodPending:
-					pending++
-				case v1.PodSucceeded:
-					succeeded++
-				case v1.PodFailed:
-					failed++
+			_, retain := podRetainPhase[pod.Status.Phase]
+
+			if !retain {
+				err := cc.deleteJobPod(job.Name, pod)
+				if err == nil {
+					terminating++
+					continue
 				}
+				// record the err, and then collect the pod info like retained pod
+				errs = append(errs, err)
+			}
+
+			switch pod.Status.Phase {
+			case v1.PodRunning:
+				running++
+			case v1.PodPending:
+				pending++
+			case v1.PodSucceeded:
+				succeeded++
+			case v1.PodFailed:
+				failed++
 			}
 		}
 	}

diff --git a/pkg/controllers/job/state/aborted.go b/pkg/controllers/job/state/aborted.go
@@ -28,12 +28,12 @@ type abortedState struct {
 func (as *abortedState) Execute(action vkv1.Action) error {
 	switch action {
 	case vkv1.ResumeJobAction:
-		return KillJob(as.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(as.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			status.State.Phase = vkv1.Restarting
 			status.RetryCount++
 			return true
 		})
 	default:
-		return KillJob(as.job, nil)
+		return KillJob(as.job, PodRetainPhaseSoft, nil)
 	}
 }
diff --git a/pkg/controllers/job/state/aborting.go b/pkg/controllers/job/state/aborting.go
@@ -31,12 +31,12 @@ func (ps *abortingState) Execute(action vkv1.Action) error {
 	switch action {
 	case vkv1.ResumeJobAction:
 		// Already in Restarting phase, just sync it
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			status.RetryCount++
 			return false
 		})
 	default:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			// If any "alive" pods, still in Aborting phase
 			if status.Terminating != 0 || status.Pending != 0 || status.Running != 0 {
 				return false

diff --git a/pkg/controllers/job/state/completing.go b/pkg/controllers/job/state/completing.go
@@ -26,7 +26,7 @@ type completingState struct {
 }
 
 func (ps *completingState) Execute(action vkv1.Action) error {
-	return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+	return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 		// If any "alive" pods, still in Completing phase
 		if status.Terminating != 0 || status.Pending != 0 || status.Running != 0 {
 			return false

diff --git a/pkg/controllers/job/state/factory.go b/pkg/controllers/job/state/factory.go
@@ -17,18 +17,28 @@ limitations under the License.
 package state
 
 import (
+	"k8s.io/api/core/v1"
+
 	vkv1 "volcano.sh/volcano/pkg/apis/batch/v1alpha1"
 	"volcano.sh/volcano/pkg/controllers/apis"
 )
 
+type PhaseMap map[v1.PodPhase]struct{}
 type UpdateStatusFn func(status *vkv1.JobStatus) (jobPhaseChanged bool)
 type ActionFn func(job *apis.JobInfo, fn UpdateStatusFn) error
+type KillActionFn func(job *apis.JobInfo, podRetainPhase PhaseMap, fn UpdateStatusFn) error
+
+var PodRetainPhaseNone = PhaseMap{}
+var PodRetainPhaseSoft = PhaseMap{
+	v1.PodSucceeded: {},
+	v1.PodFailed:    {},
+}
 
 var (
 	// SyncJob will create or delete Pods according to Job's spec.
 	SyncJob ActionFn
-	// KillJob kill all Pods of Job.
-	KillJob ActionFn
+	// KillJob kill all Pods of Job with phase not in podRetainPhase.
+	KillJob KillActionFn
 	// CreateJob will prepare to create Job.
 	CreateJob ActionFn
 )

diff --git a/pkg/controllers/job/state/finished.go b/pkg/controllers/job/state/finished.go
@@ -27,5 +27,5 @@ type finishedState struct {
 
 func (ps *finishedState) Execute(action vkv1.Action) error {
 	// In finished state, e.g. Completed, always kill the whole job.
-	return KillJob(ps.job, nil)
+	return KillJob(ps.job, PodRetainPhaseSoft, nil)
 }
diff --git a/pkg/controllers/job/state/inqueue.go b/pkg/controllers/job/state/inqueue.go
@@ -28,7 +28,7 @@ type inqueueState struct {
 func (ps *inqueueState) Execute(action vkv1.Action) error {
 	switch action {
 	case vkv1.RestartJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseNone, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Pending
 			if status.Terminating != 0 {
 				phase = vkv1.Restarting
@@ -39,7 +39,7 @@ func (ps *inqueueState) Execute(action vkv1.Action) error {
 		})
 
 	case vkv1.AbortJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Pending
 			if status.Terminating != 0 {
 				phase = vkv1.Aborting
@@ -48,7 +48,7 @@ func (ps *inqueueState) Execute(action vkv1.Action) error {
 			return true
 		})
 	case vkv1.CompleteJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Completed
 			if status.Terminating != 0 {
 				phase = vkv1.Completing

diff --git a/pkg/controllers/job/state/pending.go b/pkg/controllers/job/state/pending.go
@@ -28,7 +28,7 @@ type pendingState struct {
 func (ps *pendingState) Execute(action vkv1.Action) error {
 	switch action {
 	case vkv1.RestartJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseNone, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Pending
 			if status.Terminating != 0 {
 				phase = vkv1.Restarting
@@ -39,7 +39,7 @@ func (ps *pendingState) Execute(action vkv1.Action) error {
 		})
 
 	case vkv1.AbortJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Pending
 			if status.Terminating != 0 {
 				phase = vkv1.Aborting
@@ -48,7 +48,7 @@ func (ps *pendingState) Execute(action vkv1.Action) error {
 			return true
 		})
 	case vkv1.CompleteJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Completed
 			if status.Terminating != 0 {
 				phase = vkv1.Completing

diff --git a/pkg/controllers/job/state/restarting.go b/pkg/controllers/job/state/restarting.go
@@ -26,7 +26,7 @@ type restartingState struct {
 }
 
 func (ps *restartingState) Execute(action vkv1.Action) error {
-	return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+	return KillJob(ps.job, PodRetainPhaseNone, func(status *vkv1.JobStatus) bool {
 		// Get the maximum number of retries.
 		maxRetry := DefaultMaxRetry
 		if ps.job.Job.Spec.MaxRetry != 0 {

diff --git a/pkg/controllers/job/state/running.go b/pkg/controllers/job/state/running.go
@@ -28,7 +28,7 @@ type runningState struct {
 func (ps *runningState) Execute(action vkv1.Action) error {
 	switch action {
 	case vkv1.RestartJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseNone, func(status *vkv1.JobStatus) bool {
 			if status.Terminating != 0 {
 				status.State.Phase = vkv1.Restarting
 				status.RetryCount++
@@ -37,7 +37,7 @@ func (ps *runningState) Execute(action vkv1.Action) error {
 			return false
 		})
 	case vkv1.AbortJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			if status.Terminating != 0 {
 				status.State.Phase = vkv1.Aborting
 				return true
@@ -46,7 +46,7 @@ func (ps *runningState) Execute(action vkv1.Action) error {
 			return false
 		})
 	case vkv1.TerminateJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			if status.Terminating != 0 {
 				status.State.Phase = vkv1.Terminating
 				return true
@@ -55,7 +55,7 @@ func (ps *runningState) Execute(action vkv1.Action) error {
 			return false
 		})
 	case vkv1.CompleteJobAction:
-		return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+		return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 			phase := vkv1.Completed
 			if status.Terminating != 0 {
 				phase = vkv1.Completing

diff --git a/pkg/controllers/job/state/terminating.go b/pkg/controllers/job/state/terminating.go
@@ -26,7 +26,7 @@ type terminatingState struct {
 }
 
 func (ps *terminatingState) Execute(action vkv1.Action) error {
-	return KillJob(ps.job, func(status *vkv1.JobStatus) bool {
+	return KillJob(ps.job, PodRetainPhaseSoft, func(status *vkv1.JobStatus) bool {
 		// If any "alive" pods, still in Terminating phase
 		if status.Terminating != 0 || status.Pending != 0 || status.Running != 0 {
 			return false