[Mulgara-dev] Exception stack trace while running a backup

[Paul Gearon]

Hi David,

On May 14, 2007, at 9:14 AM, David Moll wrote:
> Great news, we have moved production off of TKS2.1 and on to  
> Mulgara!  The change was made on Tuesday, and so far it has been  
> working much better than TKS.

I'd certainly expect that to be the case!  :-)

Please let us know if you run into any issues at all.  We obviously  
want stability throughout the system, but we *really* don't want to  
go backwards wrt TKS.

<snip/>
> What’s especially interesting to me about this error is that the  
> stack trace was printed twice, once at 08:29:09, and again at  
> 08:29:24.  The only difference between the two traces is the  
> timestamp.  I am at a loss as to why this is happening.  The only  
> thing that I can think of is that an insert was attempted while the  
> backup was being processed, and the insert was finalized, causing  
> instance of the tree being backed up to be lost.  That doesn’t seem  
> likely, as I would assume that a backup process would get its own  
> phased copy of the node tree to work on while the rest of the  
> system moved forward.  Any comments or suggestions are welcome.
>
> Exception stack trace while running a backup
>
>
>
> I uploaded the stack trace to my webspace, because it won’t fit in  
> the 40k limit for this message.  Here is the link:
>
>
>
> http://www.ugcs.caltech.edu/~moll/mulgara/trace1.txt

This is an indication that release() was called twice on an AVLNode.   
We *could* note that this has already happened, and just let it go,  
but that would just be hiding the fact that it shouldn't have been  
called twice.

I've looked at the code in XAStringPoolImpl (where this problem  
arises), and the use of AVLNode seems correct.  The only thing I can  
think of is that the same AVLNode was accessed in two different  
places at once... meaning that the ObjectPool handed it out twice  
when it shouldn't have.  I have a suspicion that it's related to  
reading and writing the same nodes in different phases.

I'll keep looking at it, but I have other distractions as well.  I've  
commented on this issue in Jira at http://mulgara.org/jira/browse/ 
MGR-63 so hopefully Andrae will be able to pick it up.

Paul
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mulgara.org/pipermail/mulgara-dev/attachments/20070517/42e12bb8/attachment.htm>