Scanflow described ML workflows need to be deployed on K8s clusters. In order to improve the performance of each workflow job or service, this issue requests Scanflow could use default scheduler or enhanced volcano batch scheduler to intelligently deploy the workflows on the cluster.
Describe the solution you'd like
Use volcano batch scheduler to deploy ML workflows(Argo workflows - jobs)
Use default k8s scheduler to deploy ML services(Seldon workflows - services)
[question? how to support multi-scheduler? for current knowledge, we need to tell the scheduler which nodes can be operated.]
Policies
Support Affinity-Aware scheduling
Support proportion of resources for node
Scanflow described ML workflows need to be deployed on K8s clusters. In order to improve the performance of each workflow job or service, this issue requests Scanflow could use default scheduler or enhanced volcano batch scheduler to intelligently deploy the workflows on the cluster.
Describe the solution you'd like Use volcano batch scheduler to deploy ML workflows(Argo workflows - jobs) Use default k8s scheduler to deploy ML services(Seldon workflows - services)
[question? how to support multi-scheduler? for current knowledge, we need to tell the scheduler which nodes can be operated.]
Policies Support Affinity-Aware scheduling Support proportion of resources for node