Killing a group currently stages the killing of its group subtree by scheduling
the child groups to be killed, which when run will then recursively schedule
their own children, and so on.
Killing the entire subtree in one scheduled step is more time and space
efficient. In previous single-threaded implementations, the long pause might
have been unacceptable, but this is not an issue in our current thread pool
implementation.
Care must be taken to ensure that descent into the subtrees is not synchronized
in the parent; otherwise the entire group subtree could become synchronized in
the same thread, creating a bottleneck.
Original issue reported on code.google.com by dkitc...@gmail.com on 16 Nov 2011 at 11:29
Original issue reported on code.google.com by
dkitc...@gmail.com
on 16 Nov 2011 at 11:29