feat(deployment): gcp marketplace - provide emissary as an option. Pa…

…rt of #5718 (#6250)

manifests/gcp_marketplace/chart/kubeflow-pipelines/templates/argo.yaml

@@ -345,7 +345,8 @@ data:
   # * https://github.com/argoproj/argo-workflows/blob/v3.1.2/docs/workflow-controller-configmap.md
   # * https://github.com/argoproj/argo-workflows/blob/v3.1.2/docs/workflow-controller-configmap.yaml
 
-  containerRuntimeExecutor: docker
+  # emissary executor is a more portable default, see https://github.com/kubeflow/pipelines/issues/1654.
+  containerRuntimeExecutor: '{{ if .Values.executor.emissary }}emissary{{ else }}docker{{ end }}'
 
   # Note, {{ `some-string-{{without}}-template-interpretation` }} is a way to avoid some brackets interpreted as template.
   # Reference: https://github.com/helm/helm/issues/2798#issuecomment-467319526

manifests/gcp_marketplace/chart/kubeflow-pipelines/values.yaml

@@ -17,8 +17,9 @@ images:
   cacheserver: gcr.io/ml-pipeline/google/pipelines/cacheserver:dummy
   cachedeployer: gcr.io/ml-pipeline/google/pipelines/cachedeployer:dummy
 
+executor:
+  emissary: false
 gcpDefaultConfigName: "gcp-default-config"
-
 managedstorage:
   enabled: false
   cloudsqlInstanceConnectionName: ""

manifests/gcp_marketplace/schema.yaml

@@ -169,6 +169,30 @@ properties:
     type: string
     x-google-marketplace:
       type: NAMESPACE
+  executor.emissary:
+    type: boolean
+    title: Use emissary executor (Alpha)
+    default: false
+    description: |-
+      Kubeflow Pipelines default to use docker executor. Select this option to
+      use [the emissary executor (Alpha)](https://github.com/kubeflow/pipelines/issues/6249).
+
+      Refer to migrating to emissary executor documentation for migration instructions.
+
+      Emissary executor is more portable. On Google Kubernetes Engine (GKE)
+      clusters 1.19+, containerd is the default container runtime instead of
+      docker, so docker executor does not work properly. Emissary executor is not
+      coupled to a specific container runtime, so it works properly on any
+      Kubernetes clusters including GKE 1.19+ with containerd as runtime.
+
+      If you want to keep using docker executor for GKE 1.19+, you need to configure
+      your node pools to use docker as container runtime by selecting "Container-Optimized OS with Docker (cos)"
+      (deprecated) as the node image. For more information, refer to
+      [GKE Node Images](https://cloud.google.com/kubernetes-engine/docs/concepts/node-images).
+
+      Although emissary is in Alpha, Kubeflow Pipelines team is actively helping
+      to make it mature. Welcome any feedback if you have already tried emissary
+      executor: https://github.com/kubeflow/pipelines/issues/6249.
   managedstorage.enabled:
     type: boolean
     title: Use managed storage

manifests/gcp_marketplace/test/snapshot-base.yaml

@@ -829,7 +829,8 @@ data:
   # * https://github.com/argoproj/argo-workflows/blob/v3.1.2/docs/workflow-controller-configmap.md
   # * https://github.com/argoproj/argo-workflows/blob/v3.1.2/docs/workflow-controller-configmap.yaml
 
-  containerRuntimeExecutor: docker
+  # emissary executor is a more portable default, see https://github.com/kubeflow/pipelines/issues/1654.
+  containerRuntimeExecutor: 'docker'
 
   # Note, some-string-{{without}}-template-interpretation is a way to avoid some brackets interpreted as template.
   # Reference: https://github.com/helm/helm/issues/2798#issuecomment-467319526
@@ -2060,7 +2061,7 @@ metadata:
 spec:
   descriptor:
     type: Kubeflow Pipelines
-    version: 1.7.0-alpha.1
+    version: 1.7.0-rc.2
     description: |-
       Reusable end-to-end ML workflow
     maintainers: