This PR fixes a bug where Flagger failed to remove gateways and hosts from an Istio VirtualService when the Canary was updated to enable the delegation feature.

The Problem

When a user enables the istio route delegation by setting spec.service.delegation: true and removing gateways and hosts from the Canary spec, the reconciliation logic did not properly update the managed VirtualService.

See #1464 for more details.

The Solution

The reconciliation logic has been updated to explicitly trigger a update on the VS removing the gateways and hosts when delegation is enabled. This ensures the managed VirtualService is correctly synced.

I'm not sure if this is the best solution since at the core the problem seems to be that the reconciliation logic is doing a diff but because the virtualService.Spec (VS that is read from the cluster) is updated on the fly with empty ([]) hosts and gateways, the removal of hosts and gateways are never detected by the diff.

I thought the best option would be to simply NOT modify the virtualService variable, but this seemed to generate other bugs where a flagger.kubernetes.io/original-configuration annotation was added on the VS and the canary got stuck in progressing at 10% traffic, which I couldn't fully understand how to fix.

Testing

To validate the fix I have done the following:

• A new unit test has been added to reproduce the original bug and confirm it is now resolved.
• The existing istio delegation E2E test has been updated to cover this specific scenario, and I executed said tests on a minikube cluster.

Additional notes

This issue has some quirky workarounds that actually triggers the diff and in-turn causes the VS to be updated, for example, with the E2E I found out that setting portDiscovery: truecauses the diff logic to trigger and update the VS as expected.

Fixes #1464