Statement.cancel() no longer works with WLS 8.1?

Question

Has anyone else had a problem with the cancel() method in oracle.jdbc.OracleCallableStatement
on WLS 8.1 not doing a very good job of killing the thread in Oracle? We were
using cancel() with weblogic.jdbc.pool.CallableStatement on WLS 6.1 and it does
a very good job of killing the query on the Oracle side. But almost all of this
package has been removed and we've been told we need to use the Oracle vendor
package to get our stuff to work on 8.1. I decompiled both classes and they each
have very different implementations of the cancel() method.
I think I've managed to convert all of our DB access classes to the Oracle vendor
package successfully, with the exception of this issue. We use cancel() let users
cancel large queries and to cancel an existing query if a user tries to run a
new one. Since we implemented it we've had a huge performance boost from the lack
of runaway queries in the DB. It would be a real pain if we can't find a way to
get the Oracle version to work. Any help or tales of similar experiences would
be greatly appreciated.
thx a lot,
Matt Savino

Answer 1

I did a search on Oracle MetaLink. I didn't see any Oracle-confirmed
bugs related to cancel. I did see one user in the JDBC forum that
was having a problem with a remote (not on the local machine)
cancel. At this point, this case will need to be worked through
support - they will need to generate a reproducer and assuming they
do, file a tar with Oracle.
"Stephen Felts" <[email protected]> wrote in message
news:[email protected]...
Regarding the wrappers - now I remember discussing this with you in thebeta newsgroup about a month ago.
The goal was to make it transparent that everything was changed to bepassthrough in 8.1.
The good news is that you can now use the vendor interface directly andget their extensions
directly. That means that the wrapper for Oracle will be different fromthe wrapper for DB2.
Further, any new excentions in the interface will become visible to theapplication.
>
The bad news is that this only works when the vendor has a definedinterface; it doesn't work
if only a class is defined.
This is a problem for some of the Oracle data type classes that don't havedefined interfaces
and the reason why weblogic.vendor.oracle interfaces still need to be usedfor these classes.
That also applies to BEA classes and weblogic.jdbc.pool only had classesdefined, not interfaces.
>
We should have documented change.
Regarding the real problem of cancel(), the WLS JDBC code is not doinganything here
except passing the cancel call through to the thin driver. My guess isthat this is
a problem in the thin driver. I haven't had a chance to research this onthe Oracle site yet.
>
>
"Matt Savino" <[email protected]> wrote in message
news:[email protected]...
>>
Thanks a ton for your quick reply on this. On recommendation from yoursupport
team (case 426562) here is the primary line that we had to change:
[old] weblogic.jdbc.pool.CallableStatement cStat
=(weblogic.jdbc.pool.CallableStatement)connection.prepareCall(call);
>>
[new] - oracle.jdbc.OracleCallableStatement cStat
=(oracle.jdbc.OracleCallableStatement)connection.prepareCall(call);
>>
(FYI - We need this to take adavantage of some of the advanced Oraclefeatures
like returning multiple ResultSets. We'd like to avoid using the OCIclient if
possible, assuming that would even solve this.)
Further down in the code, here is the cancel() method:
public void cancelCall() {
try {
if (cStat != null) cStat.cancel();
releaseConnection();
catch (Exception e){ e.printStackTrace(); }
In both cases the cancel() call throws no error, but only on the oldversion do
we actually see the thread die promptly in Oracle.
Thanks again for your help on this,
Matt
"Stephen Felts" <[email protected]> wrote:
In versions prior to WLS 8.1, each Oracle extension had to be
individually,
explicitly wrapped
(and not all extensions were supported).
In WLS 8.1, it is a clean passthrough directly of all Oracle interfaces
using a dynamic proxy so that
all Oracle extensions show through. The only additional work that WLS
is doing to to
ensure that transactions are managed correctly (which shouldn't have
an impact here).
Note that in versions prior to WLS 8.1, you were using classes12.zip
for the client and now you are using ojdbc14.jar. There are
some big differences in this client implementation.
Maybe you should try testing this standalone without WLS in the picture
to see
if this is a driver problem. WLS doesn't have the code that isresponsible
for killing the thread in Oracle.
I'm not sure I understand what code you are changing. The goal was to
preseve the interfaces
provided in releases prior to 8.1. Code you show me an old code line
and what you are
changing it to? Thanks.
"Matt Savino" <[email protected]> wrote in message
news:[email protected]...
>>>>
Has anyone else had a problem with the cancel() method inoracle.jdbc.OracleCallableStatement
on WLS 8.1 not doing a very good job of killing the thread in Oracle?We were
using cancel() with weblogic.jdbc.pool.CallableStatement on WLS 6.1and it does
a very good job of killing the query on the Oracle side. But almostall of this
package has been removed and we've been told we need to use the
Oracle
vendor
package to get our stuff to work on 8.1. I decompiled both classesand they each
have very different implementations of the cancel() method.
I think I've managed to convert all of our DB access classes to theOracle vendor
package successfully, with the exception of this issue. We use
cancel()
let users
cancel large queries and to cancel an existing query if a user triesto run a
new one. Since we implemented it we've had a huge performance boostfrom the lack
of runaway queries in the DB. It would be a real pain if we can't
find
a way to
get the Oracle version to work. Any help or tales of similar
experiences
would
be greatly appreciated.
thx a lot,
Matt Savino