[FLINK-16747] Performance improvements for Python UDF

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Resolved
• Fix Version/s: 1.11.0
• Component/s: API / Python
• Assignee: Huang Xingbo
• Reporter: Dian Fu
• Created: 24/Mar/20 09:39
• Updated: 16/May/20 01:20
• Resolved: 16/May/20 01:20

Description

This is an umbrella JIRA which tracks all the efforts about Python UDF performance improvements.

Sub-Tasks

• Optimize the execution of Python UDF to use generator to eliminate unnecessary function calls: Closed
• Optimize the Python UDF execution to only serialize the value: Closed
• Change local import to global import: Closed
• Introduce FlattenRowCoder to solve the performance issue of __get_item__ in Row: Closed
• Optimize the performance of the write/read null mask in FlattenRowCoder: Closed
• Optimize the cost of function call in ScalarFunctionOpertation: Closed
• Optimize the output emit logic to remove unnecessary overhead: Closed