When changing the namespace, the workload remains in the old namespace

[ffjlabo]

What happened:

When changing the namespace, the workload remains in the old namespace.

before

% kubectl get deployment -n test simple-hoge                                                               
NAME          READY   UP-TO-DATE   AVAILABLE   AGE
simple-hoge   2/2     2            2           12m

after

% kubectl get deployment -n test2 simple-hoge                                                             
NAME          READY   UP-TO-DATE   AVAILABLE   AGE
simple-hoge   2/2     2            2           7m8s

What you expected to happen:

It would be nice to delete the old resource successfully.

How to reproduce it:

Environment:

• piped version:
• control-plane version:
• Others:

[ffjlabo]

[Root cause] Resources in the old namespace cannot be deleted because the namespace is not included in the conditions for searching the deletion target resources.

Inside PipeCD, resources are identified by four types: apiVersion, kind, namespace, and name. However, based on the live state and the manifest on git, in the pruning phase, the deletion target seems to be searched for only three things except for the namespace.

https://github.com/pipe-cd/pipecd/blob/c0d5b795de2e89ff6d3c1b7221fb2480c094ce72/pkg/app/piped/executor/kubernetes/sync.go#L131C2-L144C3

[khanhtc1202]

NOTE:

• For the same prune feature but in the config of PRIMARY_ROLLOUT we allow users to remove/prune the resource from the previous/old namespace
• This issue only happened if using the prune feature in the config of K8S_SYNC (QUICK_SYNC prune true)

=> If there is no specific reason, we should keep the spec/behavior of this prune feature the same for both PRIMARY_ROLLOUT and K8S_SYNC stage