[HIVE-16132] DataSize stats don't seem correct in semijoin opt branch

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.3.0
• Component/s: HiveServer2
• Assignee: Deepak Jaiswal
• Reporter: Deepak Jaiswal
• Created: 07/Mar/17 00:58
• Updated: 21/Jul/17 18:46
• Resolved: 13/Mar/17 18:16

Description

For the following operator tree snippet, the second Select is the start of a semijoin optimization branch. Take a look at the Data size - it is the same as the data size for its parent Select, even though the second select has only a single bigint column in its projection (the parent has 2 columns). I would expect the size to be 533328 (16 bytes * 33333).
Fixing this estimate may become important if we need to estimate the cost of generating the min/max/bloomfilter.