[SPARK-9353] Standalone scheduling memory requirement incorrect if cores per executor is not set

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.5.0
• Fix Version/s: 1.4.2, 1.5.0
• Component/s: Deploy
• Target Version/s: 1.4.2, 1.5.0
• Assignee: Andrew Or
• Reporter: Andrew Or
• Created: 26/Jul/15 06:02
• Updated: 07/Aug/15 17:29
• Resolved: 07/Aug/15 17:29

Description

I tried to come up with a more succinct title.

The issue only happens if `spark.executor.cores` is not set. Right now if we have a worker with 8G, and we set `spark.executor.memory` to 1G, then the executor launched on the worker can have at most 8 cores, even if the worker has more cores available.

This is caused by the fix in SPARK-8881.

Issue Links

• links to: [Github] Pull Request #7668 (andrewor14)
• links to: [Github] Pull Request #7686 (liancheng)