HyukjinKwon commented on issue #21977: [SPARK-25004][CORE] Add spark.executor.pyspark.memory limit. URL: https://github.com/apache/spark/pull/21977#issuecomment-455854353 I think the actual data is spilled into disk on Python RDD APIs during shuffle and/or aggregation. For instance, `partitionBy`, `sortBy`, etc. Up to my knowledge, this configuration does not apply to SQL or Arrow related APIs in Python but only RDD APIs. During aggregation and/or batch processing, it looks inevitable to hold the groups in memory and it might exceed the memory limit. In this case, it spills into disk as configured this `spark.python.worker.memory`. I think basically we should just use `spark.executor.pyspark.memory` for this if I am not mistaken here since essentially `spark.python.worker.memory` means the memory that should be used in Python worker.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
