Comments on: Merge – argh!

By: Rodrigo

Rodrigo — Fri, 28 Dec 2012 12:25:27 +0000

Jonathan/Vikas, the issue in comment 7 was fixed. We found the database server lost connection with the client due to a sort of timeout set by the firewall, but the client didn’t show any ORA error. So we enabled expire_time=10 on sqlnet.ora and the issue disappeared.
BTW, Happy New Year!!

By: Jonathan Lewis

Jonathan Lewis — Fri, 28 Dec 2012 11:21:25 +0000

Vikas,

If you’re asking about the problem reported in the main article, then comment 5 answers your question.
If you’re asking about the problem in comment 7, then I’ll have to leave it to Rodrigo to see WordPress sends him an email prompt about your question.

By: Vikas G K

Vikas G K — Fri, 28 Dec 2012 06:28:53 +0000

Hi,

We are facing the same problem in Oracle data integrator. Was this problem fixed ,if so please share the same . . .

By: Jonathan Lewis

Jonathan Lewis — Tue, 07 Aug 2012 17:38:03 +0000

Rodrigo,

In the absence of any reported error, I can’t think of any reason why the merge should roll back.

As a first step I would keep tracing the code and check the trace file for the text “ERR” – it’s possible that an error is being raised and then silently discarded; if so it should still be visible in the trace with the originating ORA- error number.

Without some idea of why this is happening there is little point in speculating whether you can work around it by putting it into a package.

By: Rodrigo

Rodrigo — Tue, 24 Jul 2012 14:50:13 +0000

Hi Jonathan,
I’m facing a very weird situation with a MERGE command and I’m quite lost. The database version is 10.2.0.4 and the OS is solaris 64 bits sparc. Basically, I have a particular MERGE command that runs for 30-40 min and suddenly starts to rollback the transaction (sometimes it runs successfully and the execution plan is exactly the same when the job fails). This is part of a sql script executed within an informatica job (it drops the indexes before the indexes, runs the MERGE and recreate the indexes), the application doesn’t fire an error and no error is seen in the alert.log. I tried to trace a 10046 event and nothing is seen, the rollback activity is also traced.
The immediate sql statement after the MERGE command is not executed when this MERGE is rolled back. While the session is rolling back , the status is of the session is ACTIVE, so this is not killed externally. I cannot trap any error from the informatica session since that tool “thinks” the session is still running.
One of the times I found an ORA 12170 in the server’s sqlnet.log, but the next time the job failed this sqlnet error was not observed.
I opened a SR in Metalink a week ago and so far they have no clue about this problem. This issue is also seen once in a while in a QA environment with a completely different network. We run several MERGE’s jobs from informatica and they don’t have any problem.
As a workaround, would it help to develop a procedure including the entire process and invoke the procedure from the client?
I know you are not a magician but you are one of the best on troubleshooting weird issues. Hope you can help me out to dig into this issue.
Thanks in advance

By: Noons

Noons — Mon, 06 Jun 2011 22:54:49 +0000

jgarr: I think it’ll be a while before 11.2.0.3 percolates to the less popular ports, such as Aix. Ah well, that delays our 11g upgade for a while more. No way I’m going to let go of a stable system and replace it with a beta release…
(no: I don’t care if I’m out of prime support, if this is what passes for prime)

By: joel garry

joel garry — Mon, 06 Jun 2011 22:41:28 +0000

That’s perfectly reasonable, it just looks scary with no qualification on 12, no way to find out when 12 will be available, and no easy way to find backports. Or am I missing some flashy MOS thing? I haven’t missed MOS 742060.1, no 11.2.0.3 there. (Excuse the grumpiness, I’m on Itanium).

By: Jonathan Lewis

Jonathan Lewis — Sun, 05 Jun 2011 11:51:02 +0000

Lars,

Interesting little oddity – the important thing now is to work out whether this is going to cause a real problem.

By: Lars

Lars — Sat, 04 Jun 2011 12:12:32 +0000

Hmm… well, if you don’t like too lazy constraint checking, maybe you prefer the over eagerness of direct path constraint checking…

create table my_test (col1 varchar(30) not null, col2 varchar(30) not null);

insert into my_test (col1) select 'x' from dual where 1=0;

–> works w/o problem

insert /*+ APPEND */ into my_test (col1) select 'x' from dual where 1=0;

ERROR at line 1:
ORA-01400: cannot insert NULL into ("MY_TEST"."COL2")

No rows are inserted, but still we get an error.

Once a default value for the second column is defined, the error is not thrown anymore…

Funny, isn’t it?

By: Jonathan Lewis

Jonathan Lewis — Sat, 04 Jun 2011 11:01:34 +0000

Sokrates,

Thanks for the follow-up.