Implicit Commit On DB2-Transaction On Termination Of Process

[Ralf Assmann]

Hi there,

we have the following problem using a DB2 version 7 on z/OS, referring
also <a
href="http://groups.google.de/group/comp.databases.ibm-db2/browse_thread/thread/2555018c0d364314/ea946c98e25ce4b7?lnk=gst&q=assmann&rnum=2#ea946c98 e25ce4b7">this
thread</a>:

Using a java program with a connection to a DB2 using a type 2-driver,
we insert many datasets into the database in one transaction. The type
2-driver connects to the database using RRSAF. If the program will be
killed (e.g. by using a kill -9 command), the program itself terminats
- but the transaction will be committed. Important is, that the program
did not send any commit command using the jdbc-connection (which cannot
be done by the program because it was killed toughly so the code
sending a rollback or commit will not be reached).

If we use a type 4-driver (which does not connect to the database using
RRSAF), everything works well. The connection will be closed doing a
rollback before. Unfortunately, it is necessary for us to use a type
2-driver, the explanation of this should not be part of this thread and
should not be discussed.

Does anyone has an idea how to solve this problem, doing a implicit
rollback closing a connection? Is there a parameter on the DB2 using
RRSAF where could be set that the database has to do a rollback while
closing a connection? A commit should only be done if the program
explicitly sends the commit-command, not at any time an implicit
commit.

We are looking forward to hearing from you soon.

Best regards,
Ralf

[Ralf Assmann]

Is there noone having an idea how to solve this problem?

Ralf

[Knut Stolze]

Ralf Assmann wrote:

Hi there,

we have the following problem using a DB2 version 7 on z/OS, referring
also <a

href="http://groups.google.de/group/comp.databases.ibm-db2/browse_thread/thread/2555018c0d364314/ea946c98e25ce4b7?lnk=gst&q=assmann&rnum=2#ea946c98 e25ce4b7">this

thread</a>:

Using a java program with a connection to a DB2 using a type 2-driver,
we insert many datasets into the database in one transaction. The type
2-driver connects to the database using RRSAF. If the program will be
killed (e.g. by using a kill -9 command), the program itself terminats
- but the transaction will be committed. Important is, that the program
did not send any commit command using the jdbc-connection (which cannot
be done by the program because it was killed toughly so the code
sending a rollback or commit will not be reached).

If we use a type 4-driver (which does not connect to the database using
RRSAF), everything works well. The connection will be closed doing a
rollback before. Unfortunately, it is necessary for us to use a type
2-driver, the explanation of this should not be part of this thread and
should not be discussed.

Does anyone has an idea how to solve this problem, doing a implicit
rollback closing a connection? Is there a parameter on the DB2 using
RRSAF where could be set that the database has to do a rollback while
closing a connection? A commit should only be done if the program
explicitly sends the commit-command, not at any time an implicit
commit.

As far as I know, that's working as designed - by the Java folks. They want
to have an implicit commit at application termination/connection close.

--
Knut Stolze
DB2 Information Integration Development
IBM Germany

[joe.weinstein]

On Nov 21, 6:07 am, Knut Stolze <[e-mail]:

Ralf Assmann wrote:
As far as I know, that's working as designed - by the Java folks. They want
to have an implicit commit at application termination/connection close.

No, it's not the Java folks, because the same Java folks made the
type-4
driver which works differently. Killing the client absolves it of any
ability
or responsibility to change the subsequent reaction of the DBMS. I
think there must be an option with the RRSAF protocol which by default
deems that a lost connection means commit, and that if the type-2
driver
is written well, there may be a connection property you can set to
alter
this behavior. The type-2 driver is broken if it is as you say. If the
client
is killed when it has done half it's intended transaction, you do *not*
want
the DBMS doing a heuristic commit.

Joe Weinstein at BEA Systems