Re: [bdbxml] Container Alias question/request...

> Version info etc:  dbxml-2.1.8 on fedora core 3 from c++ w/ gcc  
> v3.4.3.
>
> Description: I have a multi-threaded server that maintains a  
> std::map of XmlContainers that is shared across threads.  To this  
> std::map may be added or deleted XmlContainers based on the  
> requests of clients to the server.
> The pseudo logic that follows for deleting XmlContainers is the  
> basic sore spot:
>
> syncronize my_std::map
> count = my_std::map.erase( container_name );
> if ( count != 1 )
>    no such container
> else {
>  try {
>       myXmlManager.removeContainer( container_name );
>  } catch ( XmlException &e ) {
>      try { reopen container and add back to my_std::map }
>      catch (...){ log error }
>   }
> }
> release sync on map
>
> This works fine, and failures only happen if another thread has the  
> to-be-deleted container active in a long running query, etc.  This  
> makes sense and is logical behavior.
>
> The problem begins as soon as a container has an alias added for  
> it, as the XmlManager is seemingly getting a ref to the  
> XmlContainer, and when removing the container from the std::map,  
> the XmlContainer isn't the last ref, and the resource isn't closed/ 
> released, hence the XmlManager::removeContainer() will fail  
> everytime throwing ContainerOpen.

This is odd.  Adding an alias does not actually add a reference.  It
just adds another entry to the XmlManager's map of open containers.

This error may happen if you try to remove the container using the  
alias name, which
is not allowed.

>
> I know how to solve the problem, I just thought that it should be  
> doc'd that adding aliases to containers creates an additional ref  
> to the underlying container and the container doesn't go out of  
> scope as would be logically expected.  This is a gripe i guess, but  
> is it neccesary for the XmlManager to keep a ref to the actual  
> container, and not just map aliases to container names so that this  
> isn't an issue?

An alias just adds an extra *pointer* to the underlying container  
object, with no
extra reference.    So, I'm curious how you might reproduce this in a  
simple program.

George

p.s. thanks for working on nxqd!


>
> On the upside, dbxml is an absolutely fantastic product, thanks  
> much!  We will be releasing a 1.0 version of nxqd (native xml query  
> daemon) at the end of the month, so we'll keep you posted...
>
> thanks,
>
> -tom
>



------------------------------------------
To remove yourself from this list, send an
email to [email protected]