[SPARK-32953] Lower memory usage in toPandas with Arrow self_destruct

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.1
• Fix Version/s: 3.2.0
• Component/s: PySpark
• Assignee: David Li
• Reporter: David Li
• Created: 21/Sep/20 13:18
• Updated: 09/Aug/21 01:59
• Resolved: 10/Feb/21 17:59

Description

As described on the mailing list:
http://apache-spark-developers-list.1001551.n3.nabble.com/DISCUSS-Reducing-memory-usage-of-toPandas-with-Arrow-quot-self-destruct-quot-option-td30149.html https://lists.apache.org/thread.html/r581d7c82ada1c2ac3f0584615785cc60cf5ac231e1f29737d3a6569f%40%3Cdev.spark.apache.org%3E

toPandas() can as much as double memory usage as both Arrow and Pandas retain a copy of a dataframe in memory during the conversion. Arrow >= 0.16 offers a self_destruct mode that avoids this with some caveats.

Issue Links

• is related to: SPARK-34463 toPandas failed with error: buffer source array is read-only when Arrow with self-destruct is enabled
• links to: [Github] Pull Request #29818 (lidavidm)