Fix incorrect TensorBoard images. (#1109)

Signed-off-by: Syulin7 <[email protected]>
kubeflow · Jul 22, 2024 · 9064896 · 9064896
1 parent c9dbc8f
commit 9064896
Show file tree

Hide file tree

Showing 8 changed files with 11 additions and 12 deletions.
diff --git a/charts/cron-tfjob/values.yaml b/charts/cron-tfjob/values.yaml
@@ -48,7 +48,7 @@ tfjob:
   imagePullPolicy: Always
 
   useTensorboard: false
-  tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/kube-ai/tensorflow:1.5.0-devel
+  tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel
   tensorboardImagePullpolicy: Always
   tensorboardServiceType: NodePort
   tensorboardResources: {}

diff --git a/charts/etjob/values.yaml b/charts/etjob/values.yaml
@@ -16,7 +16,7 @@ shmSize: 2Gi
 privileged: false
 
 useTensorboard: false
-tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/tensorflow-samples/tensorflow:1.5.0-devel
+tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel
 tensorboardImagePullpolicy: Always
 tensorboardServiceType: NodePort
 

diff --git a/charts/mpijob/values.yaml b/charts/mpijob/values.yaml
@@ -16,7 +16,7 @@ shmSize: 2Gi
 privileged: false
 
 useTensorboard: false
-tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.5.0-devel
+tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel
 tensorboardImagePullpolicy: Always
 tensorboardServiceType: NodePort
 

diff --git a/charts/pytorchjob/values.yaml b/charts/pytorchjob/values.yaml
@@ -15,7 +15,7 @@ gitImage: registry.cn-zhangjiakou.aliyuncs.com/acs/git-sync:v3.3.5
 privileged: false
 
 useTensorboard: false
-tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/kube-ai/tensorflow:1.5.0-devel
+tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel
 tensorboardImagePullpolicy: Always
 tensorboardServiceType: NodePort
 

diff --git a/charts/tfjob/values.yaml b/charts/tfjob/values.yaml
@@ -40,7 +40,7 @@ gitImage: registry.cn-zhangjiakou.aliyuncs.com/acs/git-sync:v3.3.5
 imagePullPolicy: Always
 
 useTensorboard: false
-tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/kube-ai/tensorflow:1.5.0-devel
+tensorboardImage: registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel
 tensorboardImagePullpolicy: Always
 tensorboardServiceType: NodePort
 tensorboardResources: {}

diff --git a/pkg/apis/training/common.go b/pkg/apis/training/common.go
@@ -25,6 +25,5 @@ var DefaultCommonSubmitArgs = types.CommonSubmitArgs{
 }
 
 var DefaultSubmitTensorboardArgs = types.SubmitTensorboardArgs{
-	TensorboardImage: "registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel",
-	TrainingLogdir:   "/training_logs",
+	TrainingLogdir: "/training_logs",
 }
diff --git a/pkg/apis/types/submit_tfjob.go b/pkg/apis/types/submit_tfjob.go
@@ -107,9 +107,9 @@ type SubmitTFJobArgs struct {
 
 // SubmitTensorboardArgs is used to store tensorborad information
 type SubmitTensorboardArgs struct {
-	UseTensorboard   bool   `yaml:"useTensorboard"`   // --tensorboard
-	TensorboardImage string `yaml:"tensorboardImage"` // --tensorboardImage
-	TrainingLogdir   string `yaml:"trainingLogdir"`   // --logdir
+	UseTensorboard   bool   `yaml:"useTensorboard"`             // --tensorboard
+	TensorboardImage string `yaml:"tensorboardImage,omitempty"` // --tensorboardImage
+	TrainingLogdir   string `yaml:"trainingLogdir"`             // --logdir
 	HostLogPath      string `yaml:"hostLogPath"`
 	IsLocalLogging   bool   `yaml:"isLocalLogging"`
 }

diff --git a/pkg/argsbuilder/submit_tensorboard.go b/pkg/argsbuilder/submit_tensorboard.go
@@ -85,9 +85,9 @@ func (s *SubmitTensorboardArgsBuilder) AddCommandFlags(command *cobra.Command) {
 		s.subBuilders[name].AddCommandFlags(command)
 	}
 	command.Flags().BoolVar(&s.args.UseTensorboard, "tensorboard", false, "enable tensorboard")
-	command.Flags().StringVar(&s.args.TensorboardImage, "tensorboardImage", "registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel", "the docker image for tensorboard")
+	command.Flags().StringVar(&s.args.TensorboardImage, "tensorboardImage", "", "the docker image for tensorboard")
 	_ = command.Flags().MarkDeprecated("tensorboardImage", "please use --tensorboard-image instead")
-	command.Flags().StringVar(&s.args.TensorboardImage, "tensorboard-image", "registry.cn-zhangjiakou.aliyuncs.com/acs/tensorflow:1.12.0-devel", "the docker image for tensorboard")
+	command.Flags().StringVar(&s.args.TensorboardImage, "tensorboard-image", "", "the docker image for tensorboard")
 	command.Flags().StringVar(&s.args.TrainingLogdir, "logdir", "/training_logs", "the training logs dir, default is /training_logs")
 }