add restart policy & scheduler name for workflow pods

api/openapi-spec/swagger.json

@@ -951,10 +951,18 @@
           "description": "Resource template subtype which can run k8s resources",
           "$ref": "#/definitions/io.argoproj.workflow.v1alpha1.ResourceTemplate"
         },
+        "restartPolicy": {
+          "description": "Restart policy for all containers within all the pods. One of OnFailure or Never. Default to Never. More info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle/#restart-policy",
+          "type": "string"
+        },
         "retryStrategy": {
           "description": "RetryStrategy describes how to retry a template when it fails",
           "$ref": "#/definitions/io.argoproj.workflow.v1alpha1.RetryStrategy"
         },
+        "schedulerName": {
+          "description": "If specified, all the pods will be dispatched by specified scheduler. If not specified, all the pods will be dispatched by default scheduler.",
+          "type": "string"
+        },
         "script": {
           "description": "Script runs a portion of code against an interpreter",
           "$ref": "#/definitions/io.argoproj.workflow.v1alpha1.ScriptTemplate"

pkg/apis/workflow/v1alpha1/types.go

@@ -217,6 +217,18 @@ type Template struct {
 
 	// Tolerations to apply to workflow pods.
 	Tolerations []apiv1.Toleration `json:"tolerations,omitempty"`
+
+	// Restart policy for all containers within all the pods.
+	// One of OnFailure or Never.
+	// Default to Never.
+	// More info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle/#restart-policy
+	// +optional
+	RestartPolicy *apiv1.RestartPolicy `json:"restartPolicy,omitempty"`
+
+	// If specified, all the pods will be dispatched by specified scheduler.
+	// If not specified, all the pods will be dispatched by default scheduler.
+	// +optional
+	SchedulerName *string `json:"schedulerName,omitempty"`
 }
 
 // Inputs are the mechanism for passing parameters, artifacts, volumes from one template to another

workflow/controller/workflowpod.go

@@ -105,7 +105,6 @@ func (woc *wfOperationCtx) createWorkflowPod(nodeName string, mainCtr apiv1.Cont
 			},
 		},
 		Spec: apiv1.PodSpec{
-			RestartPolicy: apiv1.RestartPolicyNever,
 			Containers: []apiv1.Container{
 				mainCtr,
 			},
@@ -149,6 +148,16 @@ func (woc *wfOperationCtx) createWorkflowPod(nodeName string, mainCtr apiv1.Cont
 	addSchedulingConstraints(pod, wfSpec, tmpl)
 	woc.addMetadata(pod, tmpl)
 
+	if tmpl.RestartPolicy != nil {
+		pod.Spec.RestartPolicy = *tmpl.RestartPolicy
+	} else {
+		pod.Spec.RestartPolicy = apiv1.RestartPolicyNever
+	}
+
+	if tmpl.SchedulerName != nil {
+		pod.Spec.SchedulerName = *tmpl.SchedulerName
+	}
+
 	err := addVolumeReferences(pod, wfSpec, tmpl, woc.wf.Status.PersistentVolumeClaims)
 	if err != nil {
 		return nil, err

workflow/validate/validate.go

@@ -13,6 +13,7 @@ import (
 	"github.com/argoproj/argo/workflow/artifacts/hdfs"
 	"github.com/argoproj/argo/workflow/common"
 	"github.com/valyala/fasttemplate"
+	apiv1 "k8s.io/api/core/v1"
 	apivalidation "k8s.io/apimachinery/pkg/util/validation"
 )
 
@@ -237,6 +238,12 @@ func validateNonLeaf(tmpl *wfv1.Template) error {
 	if tmpl.RetryStrategy != nil {
 		return errors.Errorf(errors.CodeBadRequest, "templates.%s.retryStrategy is only valid for container templates", tmpl.Name)
 	}
+	if tmpl.RestartPolicy != nil {
+		return errors.Errorf(errors.CodeBadRequest, "templates.%s.restartPolicy is only valid for leaf templates", tmpl.Name)
+	}
+	if tmpl.SchedulerName != nil {
+		return errors.Errorf(errors.CodeBadRequest, "templates.%s.schedulerName is only valid for leaf templates", tmpl.Name)
+	}
 	return nil
 }
 
@@ -264,6 +271,12 @@ func validateLeaf(scope map[string]interface{}, tmpl *wfv1.Template) error {
 			}
 			mountPaths[art.Path] = fmt.Sprintf("inputs.artifacts.%s", art.Name)
 		}
+		if tmpl.RestartPolicy != nil {
+			if *tmpl.RestartPolicy != apiv1.RestartPolicyNever &&
+				*tmpl.RestartPolicy != apiv1.RestartPolicyOnFailure {
+				return errors.Errorf(errors.CodeBadRequest, "templates.%s.restartPolicy should be one of Never or OnFailure", tmpl.Name)
+			}
+		}
 	}
 	if tmpl.ActiveDeadlineSeconds != nil {
 		if *tmpl.ActiveDeadlineSeconds <= 0 {