[HAWQ-1450] New HAWQ executor with vectorization & possible code generation

• Type: New Feature
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: backlog, 2.4.0.0
• Component/s: Query Execution
• Assignee: Hongxu Ma
• Reporter: Lei Chang
• Created: 03/May/17 01:01
• Updated: 20/Jul/18 09:59
• Resolved: 20/Jul/18 09:59

Description

Most HAWQ executor code is inherited from postgres & gpdb. Let's discuss how to build a new hawq executor with vectorization and possibly code generation. These optimization may potentially improve the query performance a lot.

Sub-Tasks

• Add vectorized executor extension and GUC: Resolved
• Common tuple batch structure for vectorized execution: Resolved
• vectorized data types initialization and relevant function definetion: Resolved
• Vectorized execution condition check in plan tree: Resolved
• Vectorized Scan Node Framework initialization: Resolved
• Parquet table data vectorized scan: Resolved
• Vectorized Scan qualification supported: Resolved
• AO table data vectorized scan: Resolved
• add new hook api for expressions: Resolved
• Implement Vectorized Motion Node: Resolved
• refactor the vtype in order to advance the performance: Open
• implement some vectorized aggregate functions: Open
• Create Date based Vectorized type and relevant expression: Resolved
• Fix Vectorized Execution bugs which evaluate in feature test: Closed