Post recovery release a pod may cause the release of pods within the same app even they are running

I found this issue while testing recovery. It can be reproduced with the following steps:

• Create an application, it launches multiple pods, keeps them running
• Restart the scheduler, the scheduler will recover the allocations based on allocated pods
• App gets recovered, so as its pods
• Kill one of the pod

Expectation: only one pod gets released and removed from this app. But I saw: all existing allocations are released.