[ARROW-14009] [C++] Ensure SourceNode truly feeds batches to plan in parallel

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 6.0.0
• Component/s: C++
• Labels: pull-request-available query-engine
• External issue URL: https://github.com/apache/arrow/issues/29609
• Assignee: David Li
• Reporter: David Li
• Created: 15/Sep/21 21:11
• Updated: 11/Jan/23 08:36
• Resolved: 17/Sep/21 21:44

Description

While looking at some basic instrumentation, I noticed SourceNode calls InputReceived directly in its Loop handler, meaning that we only ever process one batch at a time (barring a pipeline breaker). We should instead submit batches in parallel.

Issue Links

• causes: ARROW-13997 [C++] restore exec node based query performance
• causes: ARROW-14040 [C++] Spurious test failure in ScanNode.MinimalGroupedAggEndToEnd
• links to: GitHub Pull Request #11165