[
https://issues.apache.org/jira/browse/IGNITE-8386?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16683343#comment-16683343
]
Vladimir Ozerov commented on IGNITE-8386:
-----------------------------------------
[~jooger], two comments from my side:
1) Please provide TC Bot visa confirming that there are no new failures
2) I noticed that we use different set of columns for PK index. However, we use
the same set of columns for both "old" and "new" modes for secondary indexes.
Shouldn't we unwrap secondary index columns as well? We can do that in separate
ticket, but chances that changes will be minimal, provided that we already have
all necessary infrastructure.
> SQL: Make sure PK index do not use wrapped object
> -------------------------------------------------
>
> Key: IGNITE-8386
> URL: https://issues.apache.org/jira/browse/IGNITE-8386
> Project: Ignite
> Issue Type: Task
> Components: sql
> Affects Versions: 2.4
> Reporter: Vladimir Ozerov
> Assignee: Yury Gerzhedovich
> Priority: Major
> Labels: iep-19, performance
> Fix For: 2.8
>
>
> Currently PK may be built over the whole {{_KEY}} column, i.e. the whole
> binary object. This could happen in two cases:
> 1) Composite PK
> 2) Plain PK but with {{WRAP_KEY}} option.
> This is critical performance issue for two reasons:
> 1) This index is effectively useless and cannot be used in any sensible
> queries; it just wastes space and makes updates slower
> 2) Binary object typically has common header bytes what may lead to excessive
> number of comparisons during index update.
> To mitigate the problem we need to ensure that index is *never* built over
> {{_KEY}}, Instead, we must always extract target columns and build normal
> index over them.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
