[Biojava-dev] Closing ResultSets and Statements in biosql package.

[David Huen]

On Thursday 31 Jul 2003 9:39 am, Keith James wrote:
> >>>>> "Len" == Len Trigg <len at reeltwo.com> writes:
>
>     Len> Guys, FYI - I just committed a fix to a bug where I was
>     Len> getting the following exception operating on sequences stored
>     Len> in Oracle:
>
> [...]
>
> Excellent! I had expressed concern about this before, but couldn't
> produce a real-life example. Would it be possible for you to describe
> how to recreate the exception so that we can make a regression test?
> Do you think it is possible to cause the exception using a minimal set
> of test data?
>
> I'm proposing the addition of a new test target 'biosqltests' which
> would specifically exercise the O/R mapping, safety and performance of
> the biosql code. This would probably use DBUnit to populate a test
> database as it integrates nicely with Ant/JUnit (plus I've used it
> before on Oracle and Postgres).
>
I think it will be necessary to keep that jUnit test from being executed 
with the standard tests - most installs won't have a suitable DB available 
to run a BioSQL test suite.

As for actually reliably triggering running out of Connections from failure 
to close Connections, setting the number of Connections to the minimum 
possible not to fubar on normal code and allocating lots of memory to delay 
garbage collection of dead Connections would probably be best.  Even this 
is not deterministic as some dead Connections do get finalized by the GC 
and closed thereby.  Repeatedly invoking the test at high rate perhaps 
would do it better.

Incidentally, if using MySQL connection pool datasource, don't assume it is 
pooling - it isn't as it only provides the framework for a pool 
implementation supplied elsewhere.  And calling close() under these 
circumstances does nothing of the sort.  That's how I managed to reliably 
trigger out-of-connection errors on the Gadfly bridge.  Also, when using 
the Apache connection pooling stuff with mySQL, I occasionally seem to get 
a Connection from the pool that is closed already and it appears to be 
necessary to get the Apache code to test the Connection before handing it 
out - go figure....

Regards,
DH

-- 

David Huen, Ph.D.              Email: smh1008 at cus.cam.ac.uk
Dept. of Genetics              Fax  : +44 1223 333992
University of Cambridge        Phone: +44 1223 333982/766748
Cambridge, CB2 3UH
U.K.