[HIVE-24237] Multi level/dimensional bucketing in Hive

• Type: New Feature
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 3.1.1, 3.1.2
• Component/s: Database/Schema
• Tags: bucketing
• Assignee: Unassigned
• Reporter: Pushpender Garg
• Created: 06/Oct/20 14:44
• Updated: 06/Oct/20 14:45

Description

Hive can considerably optimize the execution of certain queries like filter, aggregations, joins, if bucketed columns are used in query for these operations. Buckets can be created on multiple columns as well where hash function is computed after merging all bucket columns. 

The problem is that if buckets are created on multiple columns but query is on subset of those columns then hive doesn't optimize that query. Unless all bucket columns are used as predicate, bucketing will not be utilized. Solution proposed is to solve this problem such that even if subset of bucket columns are used still hive will be able to optimize that query.

Instead of storing data in single dimensional buckets it can be stored in multi-dimensional buckets when multiple columns are given. If subset of bucketed columns is used as predicates in query then based on hash value of individual columns, appropriate buckets can be identified and only those buckets will be scanned. This will enable optimizations even when single column or few columns are used in querying