During a long period of downtime, in-flight jobs need to be given a new chance to check in before being orphaned. When booting, all jobs that are state == running should have modified to the current time, before any orphan logic has a chance to take place.
This conflicts with desires to have multiple API instances. We'll have to be smarter about heartbeating to make that transition.
During a long period of downtime, in-flight jobs need to be given a new chance to check in before being orphaned. When booting, all jobs that are
state == running
should havemodified
to the current time, before any orphan logic has a chance to take place.This conflicts with desires to have multiple API instances. We'll have to be smarter about heartbeating to make that transition.