Though the PODs stick around, they won't be running once when the business-logic/operation in the app completes. The container/POD will switch to "Complete" status when they are done with the operation. In other words, they won't be consuming any resources. That is the current design - no other cleanup routine will be triggered.
We are thinking of optionally also providing a flag to clean up resources, so there's no residue of previous runs.
If you're observing something other than what was described above, feel free to update the description and clarify with more details. Perhaps also include the stream/task definition, and the pod-list output.