Hi Andreas, I have just done a quick code review of ObjectWrapperManager and the lack of a mutex lock on the management of the internal _wrappers and _compressors looks likely to be the central problem in it's threadsafety - in particular during the initial phase of registering of wrappers it looks like race conditions are most likely. I don't think just re-trying will address these potential race conditions.
If you put together a solution that works please post it for discussion. I have other dev work to complete today so can't look into it right away. BTW, what version of the OSG are you working on? Robert. On 14 October 2014 10:20, Andreas Roth <[email protected]> wrote: > Hi, > > I came across an issue in the ObjectWrapper::findWrapper (trunk). > - application starts, list of object wrappers is empty > - two nodes are loaded simultaneously by two database pager threads > - both nodes require the same object wrapper > - one of the two threads loads the appropriate serializer which registers > the needed object wrappers. > - this thread checks the list of object wrappers again after the plugin > was loaded. > - these other thread (which gets PREVIOUSLY_LOADED) for the serializer > plugin does not retry to find the requested object wrapper, but the wrapper > was just loaded by the first thread. > - The second thread returns with a load error because no object wrapper > could be found. > > I propose the object wrapper retries to lookup the wrapper when a plugin > was loaded or when the list of wrappers has changed. > > Cheers, > Andreas > > ------------------ > Read this topic online here: > http://forum.openscenegraph.org/viewtopic.php?p=61293#61293 > > > > > > _______________________________________________ > osg-users mailing list > [email protected] > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org >
_______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
