You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Evgeny Stanilovsky (Jira)" <ji...@apache.org> on 2021/09/07 12:48:00 UTC
[jira] [Commented] (IGNITE-13744) Calcite. Use TableSpool for
IgniteNestedLoopJoin
[ https://issues.apache.org/jira/browse/IGNITE-13744?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17411206#comment-17411206 ]
Evgeny Stanilovsky commented on IGNITE-13744:
---------------------------------------------
[~korlov] i merge new sql-calcite branch and mute some suspicious tests, i also revert my old erroneous assumptions that table scan is more convenient than index scan also with empty conditions, plz check it ?
> Calcite. Use TableSpool for IgniteNestedLoopJoin
> ------------------------------------------------
>
> Key: IGNITE-13744
> URL: https://issues.apache.org/jira/browse/IGNITE-13744
> Project: Ignite
> Issue Type: Bug
> Components: sql
> Reporter: Taras Ledkov
> Assignee: Evgeny Stanilovsky
> Priority: Minor
> Labels: calcite, calcite2-required, calcite3-required
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Now {{NestedLoopJoinNode}} uses internal buffer to save all rows of the right input.
> We have to do refactoring {{IgniteNestedLoopJoin}} to use rewind of the right input and use TableSpool for not rewindable inputs.
> This refactoring separates implementation the join logic from materialization of the right input if it is needed. In the future we can use disk offload for TableSpool etc.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)