Whether the killer greenthread will cause the database connection to leak

the heat version:0.2.12.
the database is postgres.
We found that occasional heat database connection has some long time in the "idle in transaction" state, these connections only restart the process will disappear. I think the following may be the reason:

in the file "heat\engine\service.py", The function delete_stack may kill the greenthread in some scenarios so that the connection can not return to the connection pool. For example, two green threads in the same engine are processing the same stack, one in the update, one in the delete. The updated greenthread has been fetched to the lock, then the thread may be killed by the deleting greenthread at any location to cause the database connection leak.

The attachment is an abstract test model that proves that the kill greenthread is improper that will cause the connection to leak