I have a null_resource that is used to install/uninstall Kubernetes YAML manifests, and looks as follows:

resource "null_resource" "manifest_provisioner" {
  count = var.enabled ? 1 : 0

  triggers = {
    manifest_file = <actual_content_of_the_yaml_manifest>
  }

  # Create-time provisioner
  provisioner "local-exec" {
    command = "kubectl apply -f -<<EOF\n${self.triggers.manifest_file}\nEOF"
  }

  # Destroy-time provisioner
  provisioner "local-exec" {
    when = destroy
    command = "kubectl delete -f -<<EOF\n${self.triggers.manifest_file}\nEOF"
  }
}

When adding a new manifest, or asking to delete it, the resource works as expected.

However, whenever I want to update an existing manifest, Terraform would first destroy and then create the resource, which means that an update would always incur a full deletion of Kubernetes objects which is problematic.

How can I instruct this resource to only run the create-time provisioner when updating an existing resource?