jackrabbit-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bart van der Schans (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (JCR-3263) Consistency checker performance improvements
Date Mon, 19 Mar 2012 09:19:38 GMT

    [ https://issues.apache.org/jira/browse/JCR-3263?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13232523#comment-13232523
] 

Bart van der Schans commented on JCR-3263:
------------------------------------------

Hi Thomas,

The problem is not the sequential node ids. As long as you get the node ids in the same order
as the index the performance is ok (and as long as you don't try to count them in InnoDB).
The big performance gain is from fetching a bunch of bundles in one query at once instead
of fetching a bunch of node ids and then fetch each bundle separately with a query. With the
former (in a slightly different consistency checker we created at Hippo) I can check about
a 1M nodes in three and a half minute with a batch site of 1k.

Bart



                
> Consistency checker performance improvements
> --------------------------------------------
>
>                 Key: JCR-3263
>                 URL: https://issues.apache.org/jira/browse/JCR-3263
>             Project: Jackrabbit Content Repository
>          Issue Type: Improvement
>            Reporter: Unico Hommes
>
> Currently the consistency checker loads in a batch of node ids and for each node id fetches
the corresponding bundle, its child bundles, and parent bundle separately. This makes the
consistency checker perform less than optimal and may take hours (days?) to complete for large
repositories.
> I've been able to make the checker execute about 20 times faster on my local machine
by loading in batches of node prop bundles at once. For 17000 nodes in the workspace the current
implementation ran for about 23 seconds whereas with the enhancements I made it finished in
1.2 seconds.
> Now the problem lies in the fact that loading in node prop bundles in batches may require
a lot of memory. And it is not very predictable how much per batch size because the sizes
of the individual bundles are unpredictable.
> Also the node prop bundle contains much more information than is needed for a consistency
check.
> What would be ideal in this situation is to introduce a new type - call it NodeInfo -
that contains only the structural information the checker needs to do its work. Meaning the
node id, the parent id and the child ids. In order to allow for a possible future referential
integrity check perhaps also its reference type propeties.
> The IterablePersistenceManager interface would then get an additional method:
> Map<NodeId, NodeInfo> getAllNodeInfos();
> If this is an acceptable proposal I would like to work on this and contribute a patch.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message