Ability to reconcile/rollback changes in child cluster about clusternet

[lmxia]

I pull a request on this topic like the case on kubefed, could you take a look and take more discussion. @dixudx

[panpan0000]

@dixudx mentioned:

I think the situation here might be a bit different, as the pic in this artice shows: users can exists in child cluster, so
personally i think it is reasonable to make changes in child cluster directly.

It's reasonable in Edge situation.
But there's not only day 1 but day 2 for Application LCM .
Assuming below case:

(1) in Hub Cluster, admin distributed an app to all clusters
(2) in Child Cluster, another one modified the app, example, changing its label or cpu limit
(3) In Hub Cluster, admin updated the app (e.g. from v1.0 to v2.0)
(4) the change in step 2 got lost. the final state of this app will be aligned with original label or cpu limit in hub cluster yaml.

s0, Double-Head management will be a problem.

[dixudx]

Good scenarios. 👍👍👍 @panpan0000

clusternet-agent needs to watch the resource changes and sync back the diffs.

[lmxia]

Very interesting. If it is possible, I want to on it.
Here is my thought, need more discussion.

1. deployer in clusternet-agent need a map[*metav1.APIResource] ResourceInformer to watch the resource changes
2. enqueue the namespacedname of the associated Description.

however, now we can't resolve out the Description name from the raw content in the Description.
we can Set Annotations with Description namespacedname info into the raw objet when clusternet-agent Create the raw obj.

[dixudx]

@lmxia Yes, and we also need to consider,

1. The configuration diffs between other clusters. It seems Localization could help solve this. Or a fix-named Localization could be used to help sync and aggregate all the diffs from a child cluster.
2. RBACs.
3. Others to be considered.