[SPARK-28879] Kubernetes node selector should be configurable

• Type: Improvement
• Status: In Progress
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.1.0
• Component/s: Kubernetes, Spark Core
• Assignee: Unassigned
• Reporter: Franco
• Created: 26/Aug/19 23:43
• Updated: 28/May/20 01:21

Description

Similar to SPARK-25220.

 

Having to create pod templates is not an answer because:

a) It's not available until Spark 3.0 which is unlikely to be deployed to platforms like EMR for a while as there will be major impacts to dependent libraries.

b) It is on overkill for something as fundamental as selecting different nodes for your workers as your drivers e.g. for Spot instances.

 

Would request that we add this feature in the 2.4 timeframe.

 

Issue Links

• links to: GitHub Pull Request #25589