Suresh Thalamati wrote:
By doing quick scan of the code, I think during the query optimization
Derby opens all the tables and all the indexes on each table to estimate
the costs and those containers are not closed until the nested
transaction used to compile the query is closed. Because the containers
used to estimate the cost are not closed, corresponding container files
can not be closed, so the container cache (file handle cache ) keep
growing.
I'm not very familiar with the costing mechanism of the optimizer (most of the
work I've done just assumes that the costs are correct), but I spent some time
looking at the cost estimation code and I agree with Suresh: as far as I can
tell, it looks like the optimizer opens the containers for base tables and
indexes in the query but does not close them until compilation completes.
May be it is not really required to keep all containers open
until the end of compiling the query.
It may be possible to add logic that closes and re-opens the containers during
optimization as they are needed. Note, though, that the containers are
deliberately cached in the current code and the assumption appears to be that
they will remain open throughout optimization. So I think any changes in this
area could be non-trivial and would have to take into consideration potential
performance effects. But that said, I willingly admit that I know very little
about the relevant parts of the Derby code, so I could be wrong.
In any event, maybe the best thing to do here is to file a Jira enhancement for
tracking purposes? If you (Suraj) choose to do so, the zip file containing the
reproduction could be attached to the Jira issue for ease of reference.
> But now the big problem is my application does not support jdk15. Is
> there a way we can have a fix for the above issue with using the jdk14 ?
The first step toward resolving the problem is to file a Jira issue:
http://db.apache.org/derby/DerbyBugGuidelines.html
Once that's done, the easiest way to get a fix into the codeline is to
contribute the changes yourself :) If you have the time and inclination to work
on this issue, you should feel free to join the community:
http://wiki.apache.org/db-derby/DerbyDev
Otherwise you can just file the Jira issue and then anyone who is interested can
pick it up and work to resolve it...
Army
