[IMPALA-3540] Use cost estimates to reorder compound predicates

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: Impala 2.6.0
• Component/s: Frontend
• Labels: planner ramp-up
• Target Version: Product Backlog
• Assignee: Unassigned
• Reporter: Thomas Tauber-Marshall
• Created: 13/May/16 17:56
• Updated: 05/Jun/17 18:57

Description

Building on the work of IMPALA-2805, we can use the cost and selectivity estimates to reorder conjuncts within a compound predicate, whereas currently we only reorder conjuncts that end up as different exprs.

For example, in the query:
select * from tpch.lineitem where (p_container like '%MED%' or p_size = 1) and p_partkey = 0

Current;y we are able to recognize that 'p_partkey = 0' is cheaper than '(p_container like '%MED%' or p_size = 1)' and switch their order. We could get even more of a speedup by recognizing that 'p_size = 1' is cheaper than 'p_container like '%MED&'' and also reorder them within the compound predicate.