[HIVE-15065] SimpleFetchOptimizer should decide based on metastore stats when available

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.2.0
• Fix Version/s: 2.2.0
• Component/s: Logical Optimizer
• Target Version/s: 2.2.0
• Assignee: Prasanth Jayachandran
• Reporter: Prasanth Jayachandran
• Created: 26/Oct/16 01:10
• Updated: 26/Jul/17 00:03
• Resolved: 27/Oct/16 21:23

Description

Currently the decision to use fetch optimizer or not is based on scanning the filesystem for file lengths and see if the aggregated size is less the fetch task threshold. This can be very expensive for cloud environment. This issue is mitigated to some extent by HIVE-14920 but still that requires file system scan. We can make decision based on the stats from metastore and falling back when stats is not available. Since fast stats (numRows and fileSize) is always available this should work most of the time.

Issue Links

• links to: RB