[ 
https://issues.apache.org/jira/browse/OAK-808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13648235#comment-13648235
 ] 

Lukas Eder commented on OAK-808:
--------------------------------

I had thought about this change of semantics at first, but observe closely: 
{{currentNamePathMapper()}} creates a new {{NamePathMapperImpl}} instance, 
which _already_ acts as a "proxy" to a {{documentPrefixMap}}. The 
{{documentPrefixMap}} is not materialised within the {{NamePathMapperImpl}}, 
but it is a reference to a map that is changed just as often as the original 
{{documentContext}}

In other words, changing the semantics wouldn't change the behaviour. Instead 
it would more clearly communicate it, as it already is today.
                
> Significant performance issue in TargetImportHandler's prefix mapping handling
> ------------------------------------------------------------------------------
>
>                 Key: OAK-808
>                 URL: https://issues.apache.org/jira/browse/OAK-808
>             Project: Jackrabbit Oak
>          Issue Type: Bug
>          Components: jcr
>    Affects Versions: 0.7
>            Reporter: Lukas Eder
>         Attachments: Screen Shot YourKit Profiling Session.png, 
> TargetImportHandler.java.patch
>
>
> While playing around with Oak, I have observed a significant performance 
> bottleneck in 
> {code}
> org.apache.jackrabbit.oak.jcr.xml.TargetImportHandler
> {code}
> The problem can be seen in the attached Screenshot taken from a Yourkit 
> profiling session. The TargetImportHandler contains a reference to a 
> "documentContext" ListMultimap, which acts like a stack of prefix - namespace 
> mappings. To the outside, however, it is exposed as "documentPrefixMap", a 
> regular prefix - namespace mapping map, which only references the last 
> namespace for a prefix.
> What's problematic here: "documentPrefixMap" is eagerly copied from 
> "documentContext" every time the latter is modified, which may happen a lot 
> in upstream products, such as Adobe CQ5. Instead, "documentPrefixMap" should 
> be a Map view exposing the desired data. An example implementation using 
> Guava:
> {code}
> // Create this view once, in the constructor
> this.documentPrefixMap = Maps.transformValues(documentContext.asMap(),
>     new Function<Collection<String>, String>() {
>         @Override
>         public String apply(Collection<String> input) {
>             return Iterables.getLast(input, null);
>         }
>     });
> {code}
> The above implementation might work, as "documentPrefixMap" is a read-only 
> view of the "documentContext" map. More explicit, verbose implementations may 
> prove to be even faster.
> Before I write a patch, please let me know what you think. Having a live view 
> may of course create concurrency issues, etc.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to