There's an event fired on every SQLException (see
javax.sql.ConnectionEvent) and we return the connection to the pool when
it happens.
Aaron
On Thu, 28 Sep 2000, Rickard Oberg wrote:
> Hey
>
> > available. Why? Do XADataSource implementations from database
> > vendors has their own garbage collectors?
> > I don't think so.
>
> XADataSource implementations *should* let the pool know when the connection
> has failed.
>
> > Or, maybe they hide actual Connections, ResultSets, etc. and put
> > the actual Connections back to the pool after each method invocation?
> > Impossible: AFAIK, Rickard didn't work neither in Sybase, nor in
> > Oracle :-)
>
> Uhm, correct... so..? 8-)
>
> > Assume that a Connection instance was got in the bean method and
> > then was not closed due to error, i.e. it wasn't assigned to
> > an instance field. I would expect that in this case JVM Garbage
> > Collector would collect the Connection instance and call finalize()
> > that in turn would call close().
>
> finalize() doesn't seem to be implemented in the ConnectionInPool wrapper.
> Could this be the easy fix to do?
>
> > I am not sure, but it seems to me that it is not true, since
> > the errors of this kind really cause the exhaustion of Connections.
> > Maybe the reference to the Connection instance is kept somewhere else?
>
> Nope, that's it (i.e. the ConnectionInPool wrapper).
>
> > Can anybody explain me this?
> > Assume, that I am right and some reference remains.
> > Maybe, the problem would be solved if all such referenced was replaced
> > by WeakReferences (and HashMaps by WeakHashMaps).
>
> Then we would essentially be back to the problem of "overhead per call" as
> outlined by Aaron since we have to make sure that a call invocation actually
> has a connection to work with.
>
> > Now assume that this is impossible, some strong reference is needed.
> > Then let's wrap the Connection with an individual proxy object for
> > each bean and in finalize() method of the proxy object call close()
> > for the underlying Connection.
>
> +1
>
> /Rickard
>
>
>
