For context, we are reading some configuration data (which forms the basis of our dependency tree) from a DB at runtime and using this data in conjunction with

build_reconstructable_job

to construct a job on demand. The resultant job always includes the same ops, but the execution order of the ops changes based on the configuration data we refresh from the DB. So far,

build_reconstructable_job

is the only way I've been able to create these on-demand jobs, but perhaps there is a better way I'm overlooking.