You are viewing a plain text version of this content. The canonical link for it is here.
Posted to oak-issues@jackrabbit.apache.org by "Tomek Rękawek (JIRA)" <ji...@apache.org> on 2016/11/03 14:45:59 UTC

[jira] [Commented] (OAK-3748) Heuristic method to decide if the document may be a part of the bulk update

    [ https://issues.apache.org/jira/browse/OAK-3748?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15632941#comment-15632941 ] 

Tomek Rękawek commented on OAK-3748:
------------------------------------

It seems that the optimal number of iterations for the bulk updates = 1 (docs that have already failed will fail again). Therefore can always put all the documents into the single bulk update and then re-apply failing sequentially.

There's no need for this heuristics after all, resolving as won't fix.

> Heuristic method to decide if the document may be a part of the bulk update
> ---------------------------------------------------------------------------
>
>                 Key: OAK-3748
>                 URL: https://issues.apache.org/jira/browse/OAK-3748
>             Project: Jackrabbit Oak
>          Issue Type: Improvement
>          Components: mongomk, rdbmk
>            Reporter: Tomek Rękawek
>             Fix For: 1.6
>
>         Attachments: OAK-3748.patch
>
>
> OAK-2066 introduces the bulk createOrUpdate method in the DocumentStore interface and implements it in the Mongo and RDB DSes. Both implementations try to apply the bulk changes in a few iterations. The iteration first reads the current documents and then updates them. If there's a concurrent modification between read and update, it causes a conflict. Such documents will be updated in the next iteration. The documents that can't be updated in a few trials are eventually updated in a sequential way.
> Some documents are more probable to cause a fail than other. Eg. the root document is very likely to be changed by some other process. We could create a list of such "hotspot" documents and exclude them from the bulk updates. The list can be self-maintaining, eg. documents which conflicted in more than 50% cases in the last 1h.



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