[ https://issues.apache.org/jira/browse/IGNITE-12688?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17087940#comment-17087940 ]
Ignite TC Bot commented on IGNITE-12688: ---------------------------------------- {panel:title=Branch: [pull/7685/head] Base: [master] : No blockers found!|borderStyle=dashed|borderColor=#ccc|titleBGColor=#D6F7C1}{panel} [TeamCity *--> Run :: All* Results|https://ci.ignite.apache.org/viewLog.html?buildId=5240447&buildTypeId=IgniteTests24Java8_RunAll] > Improve performance of index inline JAVA_OBJECT fields > ------------------------------------------------------ > > Key: IGNITE-12688 > URL: https://issues.apache.org/jira/browse/IGNITE-12688 > Project: Ignite > Issue Type: Bug > Reporter: Konstantin Orlov > Assignee: Konstantin Orlov > Priority: Major > Time Spent: 10m > Remaining Estimate: 0h > > Inline JAVA_OBJECT may be reason of performance drop on index creation. > The *root cause* is the low selectivity of current part of JAVA_OBJECT that > is inlined. > Now first N bytes of binary view of object is placed into inline space of the > index. > But the first offset where the two objects with the same type may be > different is 8 (HASH_CODE_POS). > With default inline = 10 we NEVER inline it: > Inline format: > 1 byte - type, > 2 bytes - size > >> 7 bytes - data; > *My proposal:* > - For metapage v4 (BPlusMetaIO) add flag *inlineObjectHash*; > - use this flag to work in compatibility mode. > - inline ONLY hash for JAVA_OBJECT fields for new indexes; > Also this approach solves the inconsistent between comparison JAVA_OBJECT by > inline and full value, because value comparator uses hash to compare object > before compare byte arrays. -- This message was sent by Atlassian Jira (v8.3.4#803005)