[IMPALA-6178] Heuristic for selecting the left-most table should consider parallelism.

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Later
• Affects Version/s: Impala 2.10.0, Impala 2.11.0
• Component/s: Frontend
• Labels: performance planner
• Target Version: Product Backlog
• Epic Color: ghx-label-7
• Assignee: Unassigned
• Reporter: Alexander Behm
• Created: 10/Nov/17 02:20
• Updated: 10/May/19 22:32
• Resolved: 10/May/19 22:32

Description

For each query block, Impala uses the estimated materialized size of tables as a heuristic for choosing the left-most table in a series of joins. Unfortunately, that heuristic does not factor in the number of hosts that tables will execute on. The number of hosts of the left-most table dictates the degree of inter-node parallelism.

To handle this tradeoff between parallelism and size we should use a heuristic like the one in IMPALA-5612.