You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@commons.apache.org by Dain Sundstrom <da...@iq80.com> on 2007/07/17 23:42:49 UTC
[DBCP] Back pointers
I'm working on a fix for the back pointers bugs DBCP-11 and DBCP-217
where the Statement.getConnection() and ResultSet.getStatement()
return the wrong objects. The fix is pretty simple; we just need to
make sure we wrap Statements and ResultSets returned from
DelegatingConnection with the matching delegating type. Anyway, I
have the fix mostly complete with a bunch of test cases, but there is
one problem...
The PerUserPoolDataSource and SharedPoolDataSource classes return the
ConnectionImpl class directly. This class is a wrapper around the
real connection so we need to wrap returned Statements which is easy
enough. The problem is these datasources use the
CPDSConnectionFactory which does not call passivate on the delegating
connection when the connection is returned to the pool so the
Statements owned by the DelegatingConnection aren't closed. To make
matters worse, CPDSConnectionFactory can't call passivate anyway
because it is in a different package and the method is protected :(
At this point I'm not sure what to do. I could fix the problem for
all DataSources except for these two, and in the future we could
rework these two to subclass PoolingDataSource. Alternative, we
could move CPDSConnectionFactory to same package as
DelegatingConnection or make is a sublcass of some ConnectionFactory
with access to the passivate method.
I really do think these datasources should be brought in line with
the main abstractions used by the other classes, but I don't think
that is something for this release (maybe for 2.0?).
Suggestion?
-dain
---------------------------------------------------------------------
To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-dev-help@jakarta.apache.org
Re: [DBCP] Back pointers
Posted by Dain Sundstrom <da...@iq80.com>.
Sounds good to me.
I attached my fix to DBCP-11. This patch assures that all all
returned Statements, PreparedStatements, CallableStatements and
ResultSets are wrapped with a delegating object, which already
properly handle the back pointers for Connection and Statement. This
patch includes an extensive test to assure that the *same* object
used to create the statement or result set is returned from either
the getConnection() or getStatementMethod().
Note we must make sure to update this test for each release of the
JDBC spec we support as new ways of obtaining these objects tend to
be added over time.
As per your suggestion below, this patch does not fix back pointers
in connections obtained from a InstanceKeyDataSource using the
cpdsadapter. The cpdsadapter does not use the delegating wrapper
internally, and will need a different solution.
DBCP-217 should also be closed.
-dain
On Jul 17, 2007, at 9:39 PM, Phil Steitz wrote:
> On 7/17/07, Dain Sundstrom <da...@iq80.com> wrote:
>> I'm working on a fix for the back pointers bugs DBCP-11 and DBCP-217
>> where the Statement.getConnection() and ResultSet.getStatement()
>> return the wrong objects. The fix is pretty simple; we just need to
>> make sure we wrap Statements and ResultSets returned from
>> DelegatingConnection with the matching delegating type.
>
> +1
>
> Anyway, I
>> have the fix mostly complete with a bunch of test cases, but there is
>> one problem...
>>
>> The PerUserPoolDataSource and SharedPoolDataSource classes return the
>> ConnectionImpl class directly. This class is a wrapper around the
>> real connection so we need to wrap returned Statements which is easy
>> enough. The problem is these datasources use the
>> CPDSConnectionFactory which does not call passivate on the delegating
>> connection when the connection is returned to the pool so the
>> Statements owned by the DelegatingConnection aren't closed. To make
>> matters worse, CPDSConnectionFactory can't call passivate anyway
>> because it is in a different package and the method is protected :(
>>
>> At this point I'm not sure what to do. I could fix the problem for
>> all DataSources except for these two, and in the future we could
>> rework these two to subclass PoolingDataSource. Alternative, we
>> could move CPDSConnectionFactory to same package as
>> DelegatingConnection or make is a sublcass of some ConnectionFactory
>> with access to the passivate method.
>>
>> I really do think these datasources should be brought in line with
>> the main abstractions used by the other classes, but I don't think
>> that is something for this release (maybe for 2.0?).
>>
>
> I think we should leave this alone for now and consider refactoring
> for 2.0, but there is a semantic difference that we need to keep in
> mind. InstanceKeyDataSource (parent of PerUser and SharedPool)
> sources connections from a ConnectionPoolDataSource. These
> datasources return connection *handles* (PooledConnection impls),
> which are not the same as DelegatingConnections. The cpdsadapter
> package is just there for older jdbc drivers that do not provide
> ConnectionPoolDataSource implementations. See the javadoc for
> InstanceKeyDataSource and also the implementation of makeObject there.
> The key difference in the contract is that InstanceKeyDataSource
> implements ConnectionEventListener, so when used with a driver that
> correctly supports ConnectionPoolDataSource, the connection handles
> handed out to users notify the pool (actually the factory in dbcp)
> when they are closed by the user. See connectionClosed in
> CPDSConnectionFactory.
>
> Phil
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-dev-help@jakarta.apache.org
Re: [DBCP] Back pointers
Posted by Phil Steitz <ph...@gmail.com>.
On 7/17/07, Dain Sundstrom <da...@iq80.com> wrote:
> I'm working on a fix for the back pointers bugs DBCP-11 and DBCP-217
> where the Statement.getConnection() and ResultSet.getStatement()
> return the wrong objects. The fix is pretty simple; we just need to
> make sure we wrap Statements and ResultSets returned from
> DelegatingConnection with the matching delegating type.
+1
Anyway, I
> have the fix mostly complete with a bunch of test cases, but there is
> one problem...
>
> The PerUserPoolDataSource and SharedPoolDataSource classes return the
> ConnectionImpl class directly. This class is a wrapper around the
> real connection so we need to wrap returned Statements which is easy
> enough. The problem is these datasources use the
> CPDSConnectionFactory which does not call passivate on the delegating
> connection when the connection is returned to the pool so the
> Statements owned by the DelegatingConnection aren't closed. To make
> matters worse, CPDSConnectionFactory can't call passivate anyway
> because it is in a different package and the method is protected :(
>
> At this point I'm not sure what to do. I could fix the problem for
> all DataSources except for these two, and in the future we could
> rework these two to subclass PoolingDataSource. Alternative, we
> could move CPDSConnectionFactory to same package as
> DelegatingConnection or make is a sublcass of some ConnectionFactory
> with access to the passivate method.
>
> I really do think these datasources should be brought in line with
> the main abstractions used by the other classes, but I don't think
> that is something for this release (maybe for 2.0?).
>
I think we should leave this alone for now and consider refactoring
for 2.0, but there is a semantic difference that we need to keep in
mind. InstanceKeyDataSource (parent of PerUser and SharedPool)
sources connections from a ConnectionPoolDataSource. These
datasources return connection *handles* (PooledConnection impls),
which are not the same as DelegatingConnections. The cpdsadapter
package is just there for older jdbc drivers that do not provide
ConnectionPoolDataSource implementations. See the javadoc for
InstanceKeyDataSource and also the implementation of makeObject there.
The key difference in the contract is that InstanceKeyDataSource
implements ConnectionEventListener, so when used with a driver that
correctly supports ConnectionPoolDataSource, the connection handles
handed out to users notify the pool (actually the factory in dbcp)
when they are closed by the user. See connectionClosed in
CPDSConnectionFactory.
Phil
---------------------------------------------------------------------
To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-dev-help@jakarta.apache.org
Re: [DBCP] Back pointers
Posted by Dain Sundstrom <da...@iq80.com>.
BTW the reason we are getting ConnectionImpls instead of raw
connections from the PooledDataSource is because
PooledConnectionImpl.getConnection():159 wraps the raw connection
with a ConnectionImpl. The line of code is commented with "the spec
requires that this return a new Connection instance". If we didn't
wrap, the back pointers would work, but we'd be violating the JDBC
spec. Once we wrap, we need to wrap all the statements and result
sets so the back pointers are valid.
Another bug, is we're not tracking statements and result sets, so
when the logical connection is closed the statements and result sets
are not closed as required buy the spec.
-dain
On Jul 17, 2007, at 2:42 PM, Dain Sundstrom wrote:
> I'm working on a fix for the back pointers bugs DBCP-11 and
> DBCP-217 where the Statement.getConnection() and
> ResultSet.getStatement() return the wrong objects. The fix is
> pretty simple; we just need to make sure we wrap Statements and
> ResultSets returned from DelegatingConnection with the matching
> delegating type. Anyway, I have the fix mostly complete with a
> bunch of test cases, but there is one problem...
>
> The PerUserPoolDataSource and SharedPoolDataSource classes return
> the ConnectionImpl class directly. This class is a wrapper around
> the real connection so we need to wrap returned Statements which is
> easy enough. The problem is these datasources use the
> CPDSConnectionFactory which does not call passivate on the
> delegating connection when the connection is returned to the pool
> so the Statements owned by the DelegatingConnection aren't closed.
> To make matters worse, CPDSConnectionFactory can't call passivate
> anyway because it is in a different package and the method is
> protected :(
>
> At this point I'm not sure what to do. I could fix the problem for
> all DataSources except for these two, and in the future we could
> rework these two to subclass PoolingDataSource. Alternative, we
> could move CPDSConnectionFactory to same package as
> DelegatingConnection or make is a sublcass of some
> ConnectionFactory with access to the passivate method.
>
> I really do think these datasources should be brought in line with
> the main abstractions used by the other classes, but I don't think
> that is something for this release (maybe for 2.0?).
>
> Suggestion?
>
> -dain
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-dev-help@jakarta.apache.org