[CALCITE-3785] Performance regression in HepPlanner#belongsToDag() due to expensive hashCode computation

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.22.0
• Labels: pull-request-available
• Assignee: Unassigned
• Reporter: Xiening Dai
• Created: 11/Feb/20 23:07
• Updated: 06/Mar/20 04:34
• Resolved: 13/Feb/20 00:26

Description

It turns out that the change in CALCITE-3713 added an extra overhead of 20~30% HepPlanner. The hash map lookup in belongsToDag becomes much slower due to expensive computation of the hashcode and equals for the list.

A simple profiling of the planner before and after CALCITE-3713 is shown below.

To test whether or not a vertex belongs to DAG, we can simply do graph.vertexSet().contains(vertex); There's no need to look up in mapDigestToVertex map, which incurs overhead of creating the map key. This problem was amplified by CALCITE-3713.

Issue Links

• links to: GitHub Pull Request #1799