[IMPALA-3987] Short-circuit evaluation of hash joins with empty build side

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: Impala 2.7.0
• Component/s: Backend
• Labels: performance
• Target Version: Product Backlog
• Assignee: Unassigned
• Reporter: Tim Armstrong
• Created: 17/Aug/16 16:06
• Updated: 25/Oct/17 00:40

Description

Many join types never return any rows if the build side is empty. In these cases we still process the whole probe input instead of short-circuiting the evaluation.

E.g. the following query unnecessarily scans the whole of lineitem.

select * 
from tpch_20_parquet.lineitem
  inner join tpch_20_parquet.orders on l_orderkey = o_orderkey
where o_comment = 'no matches for this comment';

This also affects subqueries, e.g. IN() that are rewritten to joins.

Issue Links

• is blocked by: IMPALA-3990 ExchangeNode::Close() cancel sender fragment if called before eos
• is related to: IMPALA-5789 Prune all partitions if an always-false runtime filter is received