Idea #8437
Updated by Brett Smith about 8 years ago
"on_failure gets called when an unhandled exception is raised in an actor method":https://www.pykka.org/en/latest/api/#pykka.Actor.on_failure. All our actors should use this to detect when there's an exception that means the whole process is unlikely to recover, and kill it (probably by sending SIGKILL to the process group):
* threading.ThreadError (can't create thread due to out of RAM)
* OSError reporting that we can't allocate RAM (hopefully it has errno ENOMEM? Basically we want this to replace the strategy implemented in #6321.)
* MemoryError
Define a new class that just defines this on_failure method; then have all of the other Node Manager actors use it as their superclass.