[
https://issues.apache.org/jira/browse/IGNITE-9526?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stanislav Lukyanov resolved IGNITE-9526.
----------------------------------------
Resolution: Duplicate
Closing as duplicate of IGNITE-9519 which was filed slightly earlier.
> Allow explicitly setting index inline size for Java objects
> -----------------------------------------------------------
>
> Key: IGNITE-9526
> URL: https://issues.apache.org/jira/browse/IGNITE-9526
> Project: Ignite
> Issue Type: Bug
> Components: sql
> Reporter: Stanislav Lukyanov
> Assignee: Stanislav Lukyanov
> Priority: Major
>
> Currently Ignite never inlines indexed objects when they are POJOs. It makes
> sense, as there is currently no way of inlining a POJO other than by inlining
> the binary object, and the headers of the binary objects (i.e. their first 24
> bytes) are usually the same for the same indexed field.
> However, this approach has downsides.
> Silently ignoring the specified inline size might be surprising for a user
> willing to trade (a lot of) memory for speed and setting a large inline size
> for that purpose.
> On the other hand, inlining headers and just several first bytes of POJOs in
> binary format has proven to be beneficial. It is especially beneficial in
> case page replacements are happening, as this greatly reduces the number of
> data pages required to traverse an index tree.
> The proposal is to allow POJO inlining in case the inline size is given for
> the specific POJO field. So, unlike strings and byte arrays, POJOs will not
> use the max inline size by default - as in most cases inlining POJOs isn't
> going to help.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
