Are dynamic workflow pods safe to preempt? Assumin...
# flyte-supportm
mammoth-artist-36380
05/29/2025, 2:03 PMAre dynamic workflow pods safe to preempt? Assuming they get rescheduled could preemption cause any issues?
t
thankful-minister-83577
05/29/2025, 9:56 PM
should be safe.
thankful-minister-83577
05/29/2025, 9:57 PM
shouldn’t be any issue, they should get rescheduled as normal
m
mammoth-artist-36380
05/30/2025, 6:10 PMJust so I have the right mental model: is it the case that dynamic workflows compile their subdag and then submit back to the flytepropeller to run?
t
thankful-minister-83577
06/02/2025, 5:22 PM
yes
thankful-minister-83577
06/02/2025, 5:22 PM
that’s exactly correct. it’s run-time compilation, so you have more flexibility.
thankful-minister-83577
06/02/2025, 5:23 PM
keep in mind you still can’t pass the output of a task inside the dynamic task to a python function like
rangem
mammoth-artist-36380
06/02/2025, 5:43 PMThanks, that makes sense.
Just curious, what would happen if the dynamic workflow was compiled / submitted and then a preemption happened at the tail end of the dynamic run? Would the subsequent rerun submit the same dag again?
t
thankful-minister-83577
06/04/2025, 4:03 PM
once the parent task is done the dag is safely stored on s3. if pre-emption happens before then, then yeah it’ll have to be regenerated, side-effects included.
thankful-minister-83577
06/04/2025, 4:03 PM
if after, then the dag structure will just be read from s3
m
mammoth-artist-36380
06/04/2025, 4:08 PMah nice sounds like it is "atomic", thank you for explaining!