[jira] [Comment Edited] (IGNITE-6626) SQL: Avoid materializing rows when possible

["Vladimir Ozerov (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-6626?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16235601#comment-16235601 ] 

Vladimir Ozerov edited comment on IGNITE-6626 at 11/2/17 11:44 AM:
-------------------------------------------------------------------

New test run: https://ci.ignite.apache.org/viewQueued.html?itemId=927364


was (Author: vozerov):
https://ci.ignite.apache.org/viewQueued.html?itemId=927364

> SQL: Avoid materializing rows when possible
> -------------------------------------------
>
>                 Key: IGNITE-6626
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6626
>             Project: Ignite
>          Issue Type: Task
>          Components: cache, sql
>            Reporter: Vladimir Ozerov
>            Assignee: Vladimir Ozerov
>            Priority: Major
>              Labels: performance
>             Fix For: 2.4
>
>
> We need to filter backup keys during query execution. Currently to achieve this we do the following:
> 1) Get row link
> 2) Materialize the row (!!!)
> 3) Create H2 row (H2 wrapping)
> 4) Then get key from H2 row (unwrapping)
> 5) Calculate partition through affinity function
> What it might look like:
> 1) Get row link
> 2) Get partition from link
> This ticket is to implement row filtering on B+Tree level and avoid their materialization.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)