[jira] [Commented] (OAK-1361) DocumentNodeState#compareAgainstBaseState too slow

["Michael Dürig (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/OAK-1361?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13885457#comment-13885457 ] 

Michael Dürig commented on OAK-1361:
------------------------------------

Thanks Marcel. This improves my figures a lot. 

On scenario I was looking at simulates the situation where 100 event handlers are blocked for a while and then have to catch up with 100 commits of 1000 nodes each. While not necessarily realistic the result from profiling are nevertheless interesting. I see the following stack trace coming up a lot:

{code}
"Name","Time (ms)","Samples","Level"
"org.apache.jackrabbit.oak.plugins.document.DocumentNodeState.compareAgainstBaseState(NodeState, NodeStateDiff)","188784","1302", "17"
"org.apache.jackrabbit.oak.spi.state.AbstractNodeState.getChildNodeCount(long)","178487","1242", "18"
"org.apache.jackrabbit.oak.plugins.document.DocumentNodeState$2.iterator()","172495","1202", "19"
"org.apache.jackrabbit.oak.plugins.document.DocumentNodeState$ChildNodeEntryIterator.<init>(DocumentNodeState)","172495","1202", "20"
"org.apache.jackrabbit.oak.plugins.document.DocumentNodeState$ChildNodeEntryIterator.fetchMore()","172495","1202", "21"
{code}

In fact 94% of CPU is spent in {{AbstractNodeState.getChildNodeCount(long)}}.

> DocumentNodeState#compareAgainstBaseState too slow
> --------------------------------------------------
>
>                 Key: OAK-1361
>                 URL: https://issues.apache.org/jira/browse/OAK-1361
>             Project: Jackrabbit Oak
>          Issue Type: Improvement
>          Components: mongomk
>            Reporter: Michael Dürig
>            Assignee: Marcel Reutegger
>              Labels: observation
>             Fix For: 0.16
>
>
> {{DocumentNodeState#compareAgainstBaseState}} usually falls back to the default implementation in {{AbstractNodeState#compareAgainstBaseState(NodeState, NodeStateDiff)}}, which is slow. See also the TODO in the code. This negatively affects performance when generation observation events. 



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)