Convert launcher job and statefulset worker to pod

Dockerfile

@@ -2,11 +2,11 @@ FROM golang:1.13.6 AS build
 
 ADD . /go/src/github.com/kubeflow/mpi-operator
 WORKDIR /go/src/github.com/kubeflow/mpi-operator
-RUN go build -o mpi-operator github.com/kubeflow/mpi-operator/cmd/mpi-operator.v1alpha2
+RUN make
 
 FROM gcr.io/distroless/base-debian10:latest
-COPY --from=build /go/src/github.com/kubeflow/mpi-operator/mpi-operator /opt/
+COPY --from=build /go/src/github.com/kubeflow/mpi-operator/mpi-operator.* /opt/
 COPY third_party/library/license.txt /opt/license.txt
 
-ENTRYPOINT ["/opt/mpi-operator"]
+ENTRYPOINT ["/opt/mpi-operator.v1alpha2"]
 CMD ["--help"]

cmd/mpi-operator.v1/app/server.go

@@ -155,8 +155,7 @@ func Run(opt *options.ServerOption) error {
 			kubeInformerFactory.Core().V1().ServiceAccounts(),
 			kubeInformerFactory.Rbac().V1().Roles(),
 			kubeInformerFactory.Rbac().V1().RoleBindings(),
-			kubeInformerFactory.Apps().V1().StatefulSets(),
-			kubeInformerFactory.Batch().V1().Jobs(),
+			kubeInformerFactory.Core().V1().Pods(),
 			podgroupsInformer,
 			kubeflowInformerFactory.Kubeflow().V1().MPIJobs(),
 			opt.KubectlDeliveryImage,

deploy/mpi-operator-v1.yaml → deploy/v1/mpi-operator.yaml

@@ -45,6 +45,10 @@ spec:
                     replicas:
                       type: integer
                       minimum: 1
+    additionalPrinterColumns:
+    - name: Age
+      type: date
+      jsonPath: .metadata.creationTimestamp
 ---
 apiVersion: v1
 kind: Namespace
@@ -71,9 +75,13 @@ rules:
   resources:
   - pods
   verbs:
+  - create
   - get
   - list
   - watch
+  - delete
+  - update
+  - patch
 # This is needed for the launcher Role.
 - apiGroups:
   - ""
@@ -114,24 +122,6 @@ rules:
   - list
   - update
   - watch
-- apiGroups:
-  - apps
-  resources:
-  - statefulsets
-  verbs:
-  - create
-  - list
-  - update
-  - watch
-- apiGroups:
-  - batch
-  resources:
-  - jobs
-  verbs:
-  - create
-  - list
-  - update
-  - watch
 - apiGroups:
   - apiextensions.k8s.io
   resources:
@@ -197,6 +187,7 @@ spec:
       containers:
       - name: mpi-operator
         image: mpioperator/mpi-operator:latest
+        command: ["/opt/mpi-operator.v1"]
         args: [
           "-alsologtostderr",
           "--kubectl-delivery-image",

pkg/apis/kubeflow/v1/constants.go

@@ -14,7 +14,11 @@
 
 package v1
 
+import common "github.com/kubeflow/common/pkg/apis/common/v1"
+
 const (
 	// EnvKubeflowNamespace is ENV for kubeflow namespace specified by user.
 	EnvKubeflowNamespace = "KUBEFLOW_NAMESPACE"
+	// DefaultRestartPolicy is default RestartPolicy for ReplicaSpec.
+	DefaultRestartPolicy = common.RestartPolicyNever
 )

pkg/apis/kubeflow/v1/default.go

@@ -31,16 +31,16 @@ func addDefaultingFuncs(scheme *runtime.Scheme) error {
 
 // setDefaultsTypeLauncher sets the default value to launcher.
 func setDefaultsTypeLauncher(spec *common.ReplicaSpec) {
-	if spec != nil {
-		// Only a `RestartPolicy` equal to `Never` or `OnFailure` is allowed for `Job`.
-		if spec.RestartPolicy != common.RestartPolicyNever {
-			spec.RestartPolicy = common.RestartPolicyOnFailure
-		}
+	if spec != nil && spec.RestartPolicy == "" {
+		spec.RestartPolicy = DefaultRestartPolicy
 	}
 }
 
 // setDefaultsTypeWorker sets the default value to worker.
 func setDefaultsTypeWorker(spec *common.ReplicaSpec) {
+	if spec != nil && spec.RestartPolicy == "" {
+		spec.RestartPolicy = DefaultRestartPolicy
+	}
 }
 
 func SetDefaults_MPIJob(mpiJob *MPIJob) {