How should we handle workflows changing underneath job runs

[fractaledmind]

When working locally (or even in production), it is possible to run a job (r1), have it error and stop on some step, update the code of the job changing the shape of the workflow, and then retry the job (r2). There are a number of pain-states that can occur in this situation:

• if persisted attributes were added, and the code to persist them was added to steps that occur before the run's recovery_point, that attribute will never get persisted