[
https://issues.apache.org/jira/browse/OFBIZ-5204?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13671363#comment-13671363
]
Christoph Neuroth commented on OFBIZ-5204:
------------------------------------------
Adrian, would you review my work in progress patch? I am not sure about the
best way to handle some exceptions (see TODO markers) and for now I kept the
old Notification* POJO classes and some of the getElement* methods and only
marked them as @Deprecated. I'd be happy to refactor all the code that calls
those methods to get ServiceConfigUtil completely cleaned of possible race
conditions on the Xerces obejcts, but wanted to get your feedback on the
current implementation first.
It would also be good if we could get a partial patch into svn before I start
rewriting all the callers because this patch is way too big already for my
taste (any plans to move from SVN to a DVCS? Would make contributing so much
easier...).
> Concurrent access to GenericEngineFactory causes exceptions
> -----------------------------------------------------------
>
> Key: OFBIZ-5204
> URL: https://issues.apache.org/jira/browse/OFBIZ-5204
> Project: OFBiz
> Issue Type: Bug
> Components: framework
> Affects Versions: SVN trunk
> Reporter: Christoph Neuroth
> Priority: Critical
> Attachments: GenericEngineFactoryTests.java,
> OFBIZ-5204-2013-05-31.patch, ofbiz-serviceengine.patch
>
>
> Several users (see e.g. OFBIZ-2807, OFBIZ-2793) have reported sometimes
> getting the following exception, especially during startup or after clearing
> the cache:
> {quote}Cannot find a service engine definition for the engine name [" +
> engineName + "] in the serviceengine.xml file{quote}
> By including some debug logging we found that when this happens, one or more
> of the nodes in the DOM lose attribute values, so for example this node from
> the serviceengine.xml file:
> {code}<engine name="java"
> class="org.ofbiz.service.engine.StandardJavaEngine"/>{code}
> might look like this to the calling code:
> {code}<engine name="java" class=""/>{code}
> As usual with multi-threading problems this occurs very randomly and depends
> on a lot of factors out of control, so to test the behavior we included a
> test case which emulates lots of concurrent accesses and fails most of the
> time, see GenericEngineFactoryTests.java.
> The reason for this bug is that OFBiz reuses and concurrently accesses Xerces
> DOM objects which are not thread-safe (not even for read access).
> ResourceLoader.java#getXmlDocument also states that:
> {quote}
> // This method should be avoided. DOM object trees take a lot of memory,
> so they should
> // not be cached.
> {quote}
> Therefore we would like to propose to refactor ServiceConfigUtil (and other
> places) so that it never returns Xerces dom objects and instead internally
> binds the XML to POJOs.
> We implemented an example which fixes this particular issue.
> Apply the attached patch and generate the POJOs:
> {code}
> $ patch < ofbiz-serviceengine.patch
> $ xjc ./framework/service/dtd/service-config.xsd -p
> org.ofbiz.service.config.generated -d framework/service/src
> {code}
> This fixed this particular issue for us, but we think there might be others
> issues like this wherever Document/Element objects are passed around.
> If you are okay with this implementation we could do some more work on this,
> implementing the xjc stuff in the ant build and potentially fix other places
> where the DOM objects are used as well, but we wanted to get your opinion
> first.
--
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
