Server: sched/validator: DON'T set assimilate_state by the validator.
-
The assimilator may have already started to assimilate this WU. The validator may take quite some time to chew through a chunk of workunits. Resetting the assimilate_state after that may be harmful and override the update made by the assimilator. Instead, let the transitioner decide what to do with this workunit.
-
Also too, avoid overriding a non-zero error mask that might have been set in the meantime by some other daemon
Fixes #5706
Crappy GitHub doesn't let me comment outside a hunk's context. I presume this comment needs to be removed as they contradict the new logic (and comment), I think.
Right. And I think this loop should be deleted as well, as this should be done by the transitioner then anyway.
I'm afraid the current transitioner relies on the validator setting assimilate_state as a sign of finished validation, at least under certain conditions. I need to think about this a little more.