You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Now that we have simplified our run completion events setup (#359) we can implement a controller to manage Provider custom resources (introduced in #329). The beauty of being down to one component in the provider layer (replacing our Argo Events setup with a single Deployment) is that we have no need to manage multiple resources, so we have no need for helm, and therefore the controller can apply the resource directly.
Technical Details
Permissions for the Provider controller should be added to the existing roles.yaml, so the temporary manager-providers-viewer-rolebindingClusterRoleBinding can be removed
Current Architecture
After Architecture
Acceptance Criteria
When applying a new provider custom resource, the provider service Deployment resource is created
When updating an existing Provider resource, the provider service Deployment is updated
When deleting a Provider resource, the provider service Deployment is deleted
If a provider service Deployment is updated by something other than the Provider controller, the Deployment is re-synced to the desired state specified in the Provider resource
If a provider service Deployment is deleted by something other than the Provider controller, a new provider service Deployment is created (similarly to how ReplicaSets re-create Pods when deleted)
The provider service Deployment must have a unique name (similar to how Deployments create pods with unique names)
The KFP Operator docs have been updated as needed
The text was updated successfully, but these errors were encountered:
Overview
Now that we have simplified our run completion events setup (#359) we can implement a controller to manage Provider custom resources (introduced in #329). The beauty of being down to one component in the provider layer (replacing our Argo Events setup with a single Deployment) is that we have no need to manage multiple resources, so we have no need for helm, and therefore the controller can apply the resource directly.
Technical Details
roles.yaml
, so the temporarymanager-providers-viewer-rolebinding
ClusterRoleBinding
can be removedCurrent Architecture
After Architecture
Acceptance Criteria
ReplicaSets
re-createPods
when deleted)The text was updated successfully, but these errors were encountered: