Andrey Mashenkov created IGNITE-18896:
-----------------------------------------
Summary: Sql. SortNode optimization for queries with limit.
Key: IGNITE-18896
URL: https://issues.apache.org/jira/browse/IGNITE-18896
Project: Ignite
Issue Type: Improvement
Components: sql
Reporter: Andrey Mashenkov
As of now we use 2 structures to get top-n (highest or lowest) rows from
unsorted stream.
We use PriorityQueue, which backed with heap, to hold top-n partially sorted in
backward order, and an array (as a stack) to sort and reverse to get expected
order.
Seems, we can use a single array here.
* Pass the array to the constructor of fastutil's version
ObjectHeapPriorityQueue.
* After the stream is processed, the array will be partially sorted.
* Then we can sort the array using heap-sort algo instantly, without creating a
new array.
* Return result from the end of array like we already do.
Also, I'd suggest to use ObjectArrayList.wrap() to wrap array to a Stack
interface.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
