Deprecate IsDefaultPS in TFJob CRD API

README.md

@@ -147,7 +147,6 @@ metadata:
 data:
   controller-config-file.yaml: |
     accelerators:
-      grpcServerFilePath: /opt/mlkube/grpc_tensorflow_server/grpc_tensorflow_server.py
       alpha.kubernetes.io/nvidia-gpu:
         volumes:
           - name: <volume-name> # Desired name of the volume, ex: nvidia-libs
@@ -405,8 +404,7 @@ metadata:
 spec:
   RuntimeId: 76no
   replicaSpecs:
-  - IsDefaultPS: false
-    replicas: 1
+  - replicas: 1
     template:
       metadata:
         creationTimestamp: null
@@ -418,8 +416,7 @@ spec:
         restartPolicy: OnFailure
     tfPort: 2222
     tfReplicaType: MASTER
-  - IsDefaultPS: false
-    replicas: 1
+  - replicas: 1
     template:
       metadata:
         creationTimestamp: null
@@ -431,8 +428,7 @@ spec:
         restartPolicy: OnFailure
     tfPort: 2222
     tfReplicaType: WORKER
-  - IsDefaultPS: true
-    replicas: 2
+  - replicas: 2
     template:
       metadata:
         creationTimestamp: null

build/images/tf_operator/Dockerfile

@@ -6,12 +6,11 @@ RUN mkdir -p /opt/mlkube/test
 RUN mkdir -p /opt/tensorflow_k8s/dashboard/
 COPY tf_operator /opt/mlkube
 COPY e2e /opt/mlkube/test
-COPY grpc_tensorflow_server.py /opt/mlkube/grpc_tensorflow_server/grpc_tensorflow_server.py
 COPY backend /opt/tensorflow_k8s/dashboard/
 COPY build /opt/tensorflow_k8s/dashboard/frontend/build
 RUN chmod a+x /opt/mlkube/tf_operator
 RUN chmod a+x /opt/mlkube/test/e2e
 RUN chmod a+x /opt/tensorflow_k8s/dashboard/backend
 
 # TODO(jlewi): Reduce log level.
-ENTRYPOINT ["/opt/mlkube/tf_operator", "-alsologtostderr"]
+ENTRYPOINT ["/opt/mlkube/tf_operator", "-alsologtostderr"]

build/images/tf_operator/build_and_push.py

@@ -123,8 +123,7 @@ def main():  # pylint: disable=too-many-locals, too-many-statements
       os.path.join(go_path, "bin/tf_operator"),
       os.path.join(go_path, "bin/e2e"),
       os.path.join(go_path, "bin/backend"),
-      "dashboard/frontend/build",
-      "hack/grpc_tensorflow_server/grpc_tensorflow_server.py"
+      "dashboard/frontend/build"
   ]
 
   for s in sources:

developer_guide.md

@@ -121,20 +121,10 @@ export MY_POD_NAME=my-pod
   set the corresponding namespace for the resource.
 * TODO(jlewi): Do we still need to set MY_POD_NAME? Why?
 
-Make a copy of `grpc_tensorflow_server.py` and create a config file named `controller-config-file.yaml`:
-
-```sh
-cp hack/grpc_tensorflow_server/grpc_tensorflow_server.py /tmp/grpc_tensorflow_server.py
-
-cat > /tmp/controller-config-file.yaml << EOL
-grpcServerFilePath: /tmp/grpc_tensorflow_server.py
-EOL
-```
-
 Now we are ready to run operator locally:
 
 ```sh
-tf_operator -controller_config_file=/tmp/controller_config_file.yaml
+tf_operator --logtostderr
 ```
 
 The command creates a CRD `tfjobs` and block watching for creation of the resource kind. To verify local

pkg/apis/tensorflow/v1alpha1/defaults.go

@@ -2,7 +2,6 @@ package v1alpha1
 
 import (
 	"github.com/golang/protobuf/proto"
-	"k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/runtime"
 )
 
@@ -32,11 +31,6 @@ func SetDefaults_TFJob(obj *TFJob) {
 		if r.Replicas == nil {
 			r.Replicas = proto.Int32(Replicas)
 		}
-
-		//Set the default configuration for a PS server if the user didn't specify a PodTemplateSpec
-		if r.Template == nil && r.TFReplicaType == PS {
-			setDefault_PSPodTemplateSpec(r, c.TFImage)
-		}
 	}
 	if c.TerminationPolicy == nil {
 		c.TerminationPolicy = &TerminationPolicySpec{
@@ -48,24 +42,3 @@ func SetDefaults_TFJob(obj *TFJob) {
 	}
 
 }
-
-func setDefault_PSPodTemplateSpec(r *TFReplicaSpec, tfImage string) {
-	r.IsDefaultPS = true
-	r.Template = &v1.PodTemplateSpec{
-		Spec: v1.PodSpec{
-			Containers: []v1.Container{
-				v1.Container{
-					Image: tfImage,
-					Name:  "tensorflow",
-					VolumeMounts: []v1.VolumeMount{
-						v1.VolumeMount{
-							Name:      "ps-config-volume",
-							MountPath: "/ps-server",
-						},
-					},
-				},
-			},
-			RestartPolicy: v1.RestartPolicyOnFailure,
-		},
-	}
-}

pkg/apis/tensorflow/v1alpha1/defaults_test.go

@@ -78,27 +78,9 @@ func TestSetDefaults_TFJob(t *testing.T) {
 				Spec: TFJobSpec{
 					ReplicaSpecs: []*TFReplicaSpec{
 						{
-							Replicas: proto.Int32(1),
-							TFPort:   proto.Int32(2222),
-							Template: &v1.PodTemplateSpec{
-								Spec: v1.PodSpec{
-									Containers: []v1.Container{
-										v1.Container{
-											Image: "tensorflow/tensorflow:1.3.0",
-											Name:  "tensorflow",
-											VolumeMounts: []v1.VolumeMount{
-												v1.VolumeMount{
-													Name:      "ps-config-volume",
-													MountPath: "/ps-server",
-												},
-											},
-										},
-									},
-									RestartPolicy: v1.RestartPolicyOnFailure,
-								},
-							},
+							Replicas:      proto.Int32(1),
+							TFPort:        proto.Int32(2222),
 							TFReplicaType: PS,
-							IsDefaultPS:   true,
 						},
 					},
 					TFImage: "tensorflow/tensorflow:1.3.0",

pkg/apis/tensorflow/v1alpha1/types.go

@@ -88,8 +88,6 @@ type TFReplicaSpec struct {
 	// TFPort is the port to use for TF services.
 	TFPort        *int32 `json:"tfPort,omitempty" protobuf:"varint,1,opt,name=tfPort"`
 	TFReplicaType `json:"tfReplicaType"`
-	// IsDefaultPS denotes if the parameter server should use the default grpc_tensorflow_server
-	IsDefaultPS bool
 }
 
 type TensorBoardSpec struct {

pkg/apis/tensorflow/validation/validation.go

@@ -19,7 +19,7 @@ func ValidateTFJobSpec(c *tfv1.TFJobSpec) error {
 	// Check that each replica has a TensorFlow container and a chief.
 	for _, r := range c.ReplicaSpecs {
 		found := false
-		if r.Template == nil && r.TFReplicaType != tfv1.PS {
+		if r.Template == nil {
 			return fmt.Errorf("Replica is missing Template; %v", util.Pformat(r))
 		}