[ 
https://issues.apache.org/jira/browse/IGNITE-6663?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16264494#comment-16264494
 ] 

ASF GitHub Bot commented on IGNITE-6663:
----------------------------------------

GitHub user gg-shq opened a pull request:

    https://github.com/apache/ignite/pull/3086

    IGNITE-6663: SQL: optimize primary key equality lookup

    

You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/gridgain/apache-ignite ignite-6663

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/ignite/pull/3086.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #3086
    
----
commit 292fffbbfd992441812680030ee9b0ab1d4bc74f
Author: gg-shq <kshiro...@gridgain.com>
Date:   2017-11-23T15:47:50Z

    IGNITE-6663: Implement find(..., key, key) via findOne(), not via 
find(first, last, ...). Implement equals() and hashCode() for 
GridH2SearchRowAdapter.

----


> SQL: optimize primary key equality lookup
> -----------------------------------------
>
>                 Key: IGNITE-6663
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6663
>             Project: Ignite
>          Issue Type: Task
>          Components: sql
>            Reporter: Vladimir Ozerov
>            Assignee: Kirill Shirokov
>              Labels: performance
>             Fix For: 2.4
>
>
> H2 perform every index search through {{BaseIndex.find}} method. It contains 
> both {{first}} and {{last}} rows. If condition looks like {{attr = ?}}, then 
> both bounds are the same. When this call is propagated to our {{BPlusTree}}, 
> then two index lookups occur:
> - Lower bound: {{BPlusTree#findInsertionPoint}}
> - Upper bound: {{BPlusTree.ForwardCursor#findUpperBound}}
> This is done for a reason because we do not know in advance how many elements 
> are in between the bounds, so one lookup + scan is not an option in general 
> case. But in case of PK lookup with equality condition, when we know in 
> advance that only one row will be returned, this leads to additional 
> unnecessary comparisons. 
> Suggested fix:
> 1) Make sure that all rows in {{GridH2PlainRowFactory}} has correct 
> {{equals}} implementation.
> 2) Inside {{H2TreeIndex#find}}: if this is PK index (see constructor args) 
> and {{lower.equals(upper)}}, then use {{BPlusTree.findOne}} instead of 
> {{BPlusTree.find}}.



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

Reply via email to