[jira] [Commented] (IGNITE-13030) Calcite integration. Push projections to scans and avoid reading full row when possible

["Aleksey Plekhanov (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-13030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17638312#comment-17638312 ] 

Aleksey Plekhanov commented on IGNITE-13030:
--------------------------------------------

I've implemented inline index scan. There are some limitations caused by our B-Tree cursors implementation, but in some cases it gives a good boost. For example, query like this:
{noformat}
SELECT sum(fldIdx) FROM Item{noformat}
Before the fix executed with the folowing throughput:
{noformat}
JmhSqlBenchmark.querySumIndexed   CALCITE  thrpt    3  65,816 ±  4,518  ops/s{noformat}
After the fix:
{noformat}
JmhSqlBenchmark.querySumIndexed   CALCITE  thrpt    3  170,528 ±  8,144  ops/s{noformat}

> Calcite integration. Push projections to scans and avoid reading full row when possible
> ---------------------------------------------------------------------------------------
>
>                 Key: IGNITE-13030
>                 URL: https://issues.apache.org/jira/browse/IGNITE-13030
>             Project: Ignite
>          Issue Type: Improvement
>          Components: sql
>            Reporter: Roman Kondakov
>            Assignee: Aleksey Plekhanov
>            Priority: Minor
>              Labels: calcite2-required, calcite3-required
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> We can establish index-only scans in some cases. For example, if the column {{depId}} is indexed (and totally inlined), we can to execute a query like:
> {code:java}
> SELECT depId FROM emp
> {code}
> and collects resultset without pagememory lookups.
>  



--
This message was sent by Atlassian Jira
(v8.20.10#820010)