Hi All, I was wondering is someone could help me identify why some of my tasks are pending. I have the following flow running on prefect cloud with a dask backend. for some reason, the task scheduler has not started 2/4 of the tasks. thanks for any advice
k
Kevin Kho
04/16/2021, 2:22 AM
Hi @Vincent, what RunConfig and Executor are you using?
v
Vincent
04/16/2021, 2:23 AM
KubernetesRun with a DaskExecutor
k
Kevin Kho
04/16/2021, 2:27 AM
Did you configure the resources? This seems like the resources are limited. This thread may give ideas.
v
Vincent
04/16/2021, 2:29 AM
Indeed I did set the resource limits for the scheduler to be pretty beefy.
k
Kevin Kho
04/16/2021, 2:34 AM
I’d have to ask some team members who know more about k8s. I’ll respond to you tomorrow.
z
Zanie
04/16/2021, 1:52 PM
Hey @Vincent -- are they hanging indefinitely? Are you using timeouts? Are you returning any values from the functions that may not be serializable or use a lot of memory? Is your
DaskExecutor
being created with any arguments?
v
Vincent
04/16/2021, 3:42 PM
The tasks will usually proceed if there is some stimulus. ie. another task finishes. However, this is undesirable in the case where 1 task can hold up the entire flow. I do not specify any special args other than the cluster_class and adaptive_kwargs. The return of each function is a small and serializable, (ie. a file path).
Vincent
04/16/2021, 3:43 PM
Restarting a flow has been my workaround so far but not a long term solution.
z
Zanie
04/16/2021, 5:08 PM
Hmm. What are your cluster class/adaptation args?
Zanie
04/16/2021, 5:09 PM
And to clarify, the tasks are in a 'Pending' state?