Ray is a unified framework for scaling AI and Python applications. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
Currently pg is cleaned up after the job finishes or via a manual remote_placement_group call. That means, unlike actor, it won't be freed after it goes out of scope.
This has caused pg leaks when people failed to manually call remote_placement_group: for example a task creates a pg and that task crashes.
Description
Currently pg is cleaned up after the job finishes or via a manual
remote_placement_group
call. That means, unlike actor, it won't be freed after it goes out of scope.This has caused pg leaks when people failed to manually call
remote_placement_group
: for example a task creates a pg and that task crashes.