[runtime] Exclude default values and empty fields during resource adoption when using the annotation approach

[adriananeci]

Is your feature request related to a problem?

When adopting an S3 bucket using the annotation based mechanism(services.k8s.aws/adoption-fields and services.k8s.aws/adoption-policy: adopt annotations), the ACK object's spec is populated with all available options even if many of them are empty or have the default values.

Is it possible to populate the spec with only non-default and non-empty fields during adoption?

Describe the solution you'd like
A description of what you want to happen.

Describe alternatives you've considered
A description of any alternative solutions or features you've considered.

[michaelhtm]

This issue seems to be happening specifically in s3. Will have a patch for it soon

[michaelhtm]

Hey @adriananeci, after looking into this further, we noticed that this could cause some inconsistencies between the different controllers.
It's not very straight forward figuring out what are considered empty/default fields (eg, integers, booleans, etc.)
Also i just wanted to understand if there are any issues with having empty fields..

[adriananeci]

Hey @michaelhtm,

We currently create ACK objects via some helm charts and have objects defined to something like

apiVersion: s3.services.k8s.aws/v1alpha1
kind: Bucket
metadata:
  name: {{ $bucketName }}
  namespace: {{ $namespace }}
  labels:
  {{- include "storage.labels" . | nindent 4 }}
  annotations:
    services.k8s.aws/region: {{ .Values.global.location }}
{{- if .Values.storage.adoptExistingResources }}
    services.k8s.aws/adoption-fields: |
      {
        "name": "{{ $bucketName }}"
      }
    services.k8s.aws/adoption-policy: adopt
{{- else }}
spec:
  name: {{ $bucketName }}
  lifecycle:
    rules:
    - abortIncompleteMultipartUpload:
        daysAfterInitiation: {{ .Values.storage.daysAfterInitiation }}
      id: "delete expired multi part uploads"
      status: "Enabled"
      prefix: ""
  publicAccessBlock: {{ .Values.storage.publicAccessBlock | toYaml | nindent 4 }}
  tagging:
    tagSet:
    - key: Name
      value: {{ $bucketName }}
    - key: AckName
      value: {{ $bucketName }}
    - key: kubernetes.io/cluster/{{ $clusterFullName }}
      value: owned
    {{- include "storage.tags" . | nindent 4 -}}
{{- end }}

We are planning to adopt some existing AWS resources by leveraging the adoption annotation feature and then continue to reconcile/adjust the adopted ACK resource via the helm charts.

In order to make sure we are not missing any ACK objects setting in their spec definition, I want do a diff/comparison between the ACK object spec as populated by the controller and the ACK object spec defined in helm template.

Given currently there are many fields that are empty or have a default value when an ACK object is adopted, it is pretty hard to identify if there is a gap between the object template as defined into the helm chart and the one populated during the adoption during the diff process.

For an adopted S3 bucket in particular, I've noticed many fields that are empty or have default values, like:

spec:
  accelerate: {}
  acl: private|bucket-owner-read|bucket-owner-full-control
  cors: {}
  encryption:
    rules:
    - applyServerSideEncryptionByDefault:
        sseAlgorithm: AES256
      bucketKeyEnabled: false
  grantFullControl: id=xxxxx
  grantRead: ""
  grantReadACP: ""
  grantWrite: ""
  grantWriteACP: ""
  lifecycle: {}
  logging: {}
  notification: {}
  ownershipControls:
    rules:
    - objectOwnership: BucketOwnerEnforced
  requestPayment:
    payer: BucketOwner

In order to replicate it, you can manually create an S3 bucket with the default settings via the AWS console or aws cli and then try to adopt it. Then create a new ACK S3 bucket object with only the mandatory fields (e.g. name) , and you'll notice the diff in terms of spec fields between these 2 ACK objects, even if the AWS created resources have the same settings.

[michaelhtm]

Hey @adriananeci, I attempted solving this issue.
There's a PR ready that is able to remove all empty fields when adopting an s3 bucket, but i do not see a way we can do the same for default fields.
Here's an example:

spec:
  acl: private|bucket-owner-read|bucket-owner-full-control
  encryption:
    rules:
    - applyServerSideEncryptionByDefault:
        sseAlgorithm: AES256
      bucketKeyEnabled: false
  grantFullControl: id=xxxxx
  name: bucket
  ownershipControls:
    rules:
    - objectOwnership: BucketOwnerEnforced
  requestPayment:
    payer: BucketOwner

There is no way for the controller to know which values per field are defaulted by the Service.

Removing the defaulted fields from the resource will still reconcile successfully.