pan3793 commented on PR #51948: URL: https://github.com/apache/spark/pull/51948#issuecomment-3922015524
Since @mridulm brings a valid point - this feature is also applicable to other resource managers(likely not benefit K8s, but benefits Local, Standalone, and YARN), we should not make the config name specific to yarn. Currently, Spark has the following configs for the cores. ``` spark.yarn.am.cores spark.driver.cores spark.executor.cores ``` I wonder if we should add counterparts of this feature for them (comments/suggestions are welcome) ``` spark.yarn.am.limitActiveProcessorCount.enabled spark.driver.limitActiveProcessorCount.enabled spark.executor.limitActiveProcessorCount.enabled ``` For implementation, I think this PR modifies the right place - Given that Spark supports `ResourceProfile`, which means that executor cores might be variant in a single app, we can NOT apply `ActiveProcessorCount` for executor like `SparkContext.supplementJavaModuleOptions|supplementJavaIPv6Options`, instead, we should apply the `ActiveProcessorCount` on assembling the command for driver/executor bootstrap, where we know the cores/memory exactly. @mridulm @jzhuge WDYT? (we'd better reach a consensus about the direction before moving forward) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
