Hello! We have occasionally come across issues with flyte worker nodes crashing in EKS, and it always gets resolved by adding

concurreny

on

map_task

. (i.e.

map_task(task_function=my_task, concurrency=20)(my_input=input)

. Could someone help identify what could be happening? What is the default concurrency behavior when not set? Is there is means to configure this globally? Even concurrency values that are larger than the cluster can handle seem to resolve, implying its not purely a resourcing issue. Any help or insights are appreciated!

do you think you can repro and paste the error here?