[SPARK-37649] Switch default index to distributed-sequence by default in pandas API on Spark

• Type: Improvement
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 3.3.0
• Fix Version/s: 3.3.0
• Component/s: PySpark
• Labels: release-notes
• Assignee: Hyukjin Kwon
• Reporter: Hyukjin Kwon
• Created: 15/Dec/21 00:30
• Updated: 12/Dec/22 18:10
• Resolved: 15/Dec/21 02:23

Description

pandas API on Spark currently sets compute.default_index_type to sequence which relies on sending all data to one executor that easily causes OOM.

We should better switch to distributed-sequence type that truly distributes the data.

With this change, we can now leverage https://issues.apache.org/jira/browse/SPARK-36559 and https://issues.apache.org/jira/browse/SPARK-36338 by default, and end users will benefit a lot of performance improvement.

Issue Links

• links to: [Github] Pull Request #34902 (HyukjinKwon)