[C++] Parallelize execution of ScalarAggregateFunction

[asfimport]

After ARROW-8972, we have a generic chunk-based executor for aggregate functions. It should be relatively straightforward now to parallelize the execution loop of the executor.

Reporter: Wes McKinney / @wesm

Related issues:

• [C++] Implement Hash Aggregation query execution node (is blocked by)
• [C++] C++ array kernels framework and execution buildout (umbrella issue) (is a child of)
• [C++] Mean kernel aggregate (is related to)
• [C++] Incremental Variance, Standard Deviation aggregators (is related to)
• [C++] Parallelize execution of arrow::compute::ScalarFunction (is related to)
• [C++] Abstract aggregation kernel API (is related to)

_{Note: This issue was originally created as ARROW-3120. Please see the migration documentation for further details.}

[asfimport]

Wes McKinney / @wesm:
Removed from 0.13. The current Sum kernel implementation is not stateful. I suggest we wait to determine how parallel execution / synchronization will be managed for now (until we have more aggregations implemented, and begin to develop a runtime execution environment)

[asfimport]

Antoine Pitrou / @pitrou:
cc @lidavidm. It's still not obvious to me whether parallelization should occur inside the kernel machinery, rather than be orchestrated from the outside.

[asfimport]

David Li / @lidavidm:
With the current implementations of ExecNodes, it seems we are leaning towards the latter for now (e.g. ScalarAggregateNode manually maintains one state per thread and merges them at the end). That might change if and when we have a work-stealing scheduler as described on the mailing list.

[github-actions]

This issue has been marked as stale because it has had no activity in the past 365 days. Please remove the stale label or comment below, or this issue will be closed in 14 days. If this improvement is still desired but has no current owner, please add the 'Status: needs champion' label.