CLOUDP-358677: Add Flex e2e2 test

[josvazg]

Summary

Add a e2e2 test for Flex Clusters.

Proof of Work

Tested locally.

$ time make e2e2 label=flexcluster-ctlr EXPERIMENTAL=1
...
Ran 2 of 18 Specs in 224.121 seconds
SUCCESS! -- 2 Passed | 0 Failed | 0 Pending | 16 Skipped
PASS
Ginkgo ran 1 suite in 3m59.683338625s
Test Suite Passed
make e2e2 label=flexcluster-ctlr EXPERIMENTAL=1  39.52s user 34.56s system 25% cpu 4:46.86 total

Checklist

• Have you linked a jira ticket and/or is the ticket in the title?
• Have you checked whether your jira ticket required DOCSP changes?
• Have you signed our CLA?

[josvazg]

test is not passing yet, needs debugging

[josvazg]

Something is still wrong with this, I will debug further

[s-urbaniak]

This is doable in this case but not all objects have Group IDs (i.e. DatabaseUsers). I rather suggest blocking deletion of the Group in the group reconciler in the first place:

func (h *Handlerv20250312) HandleDeletionRequested(ctx context.Context, group *akov2generated.Group) (ctrlstate.Result, error) {
	var dependents []reconcile.Request
	dependents = append(dependents, indexers.NewFlexClusterByGroupMapFunc(h.kubeClient)(ctx, group)...)
	dependents = append(dependents, indexers.NewDatabaseUserByGroupMapFunc(h.kubeClient)(ctx, group)...)
	dependents = append(dependents, indexers.NewClusterByGroupMapFunc(h.kubeClient)(ctx, group)...)
	if len(dependends) > 0 {
		return result.NextState(state.StateDeletionRequested, fmt.Sprintf("failed to delete group because %v resources depend on it.", len(dependents)))
	}

	if customresource.IsResourcePolicyKeepOrDefault(group, h.deletionProtection) {
		return result.NextState(state.StateDeleted, "Group deleted.")
	}

[s-urbaniak]

The good news is that we have the mapping functions auto-generated already!

[s-urbaniak]

I just tried the above code and it works nicely. I first:

1. Tried to delete the group. This blocks because a Cluster, a FlexCluster, and a DatabaseUser depend on it:

2. Deleted the Cluster, a FlexCluster, and a DatabaseUser. In the meantime the count of dependend resources went down:

3. Finally the deletion in Atlas was initiated and executed:

[s-urbaniak]

side note: this is a pretty canonical use of indexes in the combination with finalizers (in our case nicely gated via the universal state machine)

[s-urbaniak]

I rather suggest blocking the group vs. having smartness in the reconciler as not all resources have a groupId in their status, i.e. DatabaseUsers