jackrabbit-oak-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chetan Mehrotra (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (OAK-5212) Avoid updating the index nodestate if no change is done in index
Date Mon, 05 Dec 2016 13:19:58 GMT

    [ https://issues.apache.org/jira/browse/OAK-5212?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15722244#comment-15722244
] 

Chetan Mehrotra commented on OAK-5212:
--------------------------------------

Thanks Vikas and Tommaso for the feedback.

bq.  but I think the DirectoryReader#listCommits is slightly more expensive as it requires
IO (read)

Makes sense. I have improved the approach a bit to detect if some write has happened via updateDocument
call. In such case the generation based check is avoided. As due to CoW reads here are local
I would apply the patch as savings on remote calls done to DocumentNodeStore would be good
enough.

> Avoid updating the index nodestate if no change is done in index
> ----------------------------------------------------------------
>
>                 Key: OAK-5212
>                 URL: https://issues.apache.org/jira/browse/OAK-5212
>             Project: Jackrabbit Oak
>          Issue Type: Improvement
>          Components: lucene
>            Reporter: Chetan Mehrotra
>            Assignee: Chetan Mehrotra
>            Priority: Minor
>              Labels: performance
>             Fix For: 1.6
>
>         Attachments: OAK-5212-v1.patch
>
>
> As noted in OAK-5211 directory listing was getting modified (due to reorder) even if
no change happens in index. 
> Another place where we update state post index close is at ":status" node where we store
{{lastUpdated}} and {{indexedNodes}} post index close. In normal cases LuceneIndexEditor avoids
initializing the IndexWriter if there is no change. However it can happen that when any node
gets deleted the editor performs a delete operation. It can happen that tree being deleted
is not indexed but still editor would do this as it cannot determine that easily. And in doing
that IndexWriter would be initialized.
> Currently IndexWriter being initialized is considered same as index updated. Due to this
index status nodes gets unnecessarily updated even if there is no change in index which causes
the IndexTracker to reopen the index even when it has not changed. 
> We should make this more explicit and find a way to determine if index has been updated
or not



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message