[SPARK-42260] Log when the K8s Exec Pods Allocator Stalls

• Type: Improvement
• Status: In Progress
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 3.4.0, 3.4.1
• Component/s: Kubernetes
• Target Version/s: 4.0.0
• Assignee: Holden Karau
• Reporter: Holden Karau
• Created: 31/Jan/23 18:34
• Updated: 09/Aug/23 19:15

Description

Sometimes if the K8s APIs are being slow the ExecutorPods allocator can stall and it would be good for us to log this (and how long we've stalled for) so folks can tell more clearly why Spark is unable to reach the desired target number of executors.

 

This is somewhat related to SPARK-36664 which logs the time spent waiting for executor allocation but goes a step further for K8s and logs when we've stalled because we have too many pending pods.

Issue Links

• links to: [Github] Pull Request #39825 (holdenk)