The task-topology plugin does not implement DeallocateFunc, which may cause the scheduling result to be unexpected

[liuyuanchun11]

Description

The plugin task-topology implements the AllocateFunc (resource allocation function) but does not implement the DeallocateFunc (resource deallocation function). If a statement.Discard operation occurs during the scheduling process, the actions performed in allocateFunc cannot be rolled back. When there are multiple pending jobs to be scheduled within a single openSession, this may lead to scheduling results that deviate from expectations.

Steps to reproduce the issue

1.Enable the task-topology plug-in
2.Create two jobs: Job-A contains 2 pods, but the current cluster's available resources can only fulfill the request for 1 pod. Job-B contains 1 pod, and the available resources are sufficient to satisfy Job-B's request.Job-A has a higher priority than Job-B.
3.After scheduling begins, Job-A first schedules one pod. However, since the cluster resources are insufficient to fulfill the second pod's request, this triggers a statement.Discard operation to roll back the allocation. Because the DeallocateFunc is not implemented, the JobManager.TaskBound state (or resource binding) for Job-A is not rolled back. This incomplete rollback may block or interfere with the subsequent scheduling of Job-B, even though Job-B's resource requirements could otherwise be met.

Describe the results you received and expected

The task-topology plugin needs to be implemented in addition to deallocate

What version of Volcano are you using?

volcano 1.10

Any other relevant information

No response