Ticket #94 (new defect)
provide control of parallel and distributed jobs
Reported by: | mmckerns | Owned by: | mmckerns |
---|---|---|---|
Priority: | major | Milestone: | mystic-dev |
Component: | pathos | Keywords: | |
Cc: |
Description (last modified by mmckerns) (diff)
provide control of parallel and distributed jobs
currently the handling mechanism fails
Change History
comment:2 Changed 6 years ago by mmckerns
from ticket #70 -- "signal handler doesn't work in parallel DE solver. Regardless of whether signal handler is enabled or disabled, a Ctrl-C will kill the execution."
comment:3 Changed 6 years ago by mmckerns
extending the Sow to log __call__(x, y, id=None) allows an automated way to 'id' each member of the map (for a particular iteration). Thus the call to StepMonitors? should take StepMonitor?(self.bestSolution[:], self.bestEnergy, self.id).
Solvers will have self.id = None by default... however for abstract_map_solvers (or abstract_nested_solvers ?), the self.id for a nested solver is set to the positional id within the map funciton. Hence, when an optimization is run on a particular node, the nested solver's self.id will log a unique id for all iterations of the nested solver (during the map).
This solution will impact the existing log readers as well (because "iteration" should be logged as the tuple of "iteration" and "id").
implementation of monitors, termination, and "condition" that are not as closely tied to solvers are being developed by Paul Kienzle in an alternate branch of mystic