DSXchange

Since two days we are getting "Abnormal termination of stage <stagename> detected" for all our jobs that read or write to Oracle oci stages.
When we reset the job, we get additional info like
From previous run
DataStage Job 15 Phantom 29653
jobnotify: Error 911 occurred.
DataStage Phantom Finished.
[110486] DSD.StageRun StagingAGTData.
StagingAGTData.SelecteerEnJoin 1 0/50 - core dumped.

This goes wrong in all our projects for any Oracle-table as long as in the job the NLS-settings are set to "project-default (NONE)". When we change the NLS of a job to another NLS, even to NONE(!), it works correctly. The same when the NLS-default of the project is changed from NONE to another NLS.

This problem occurs in every project though, including projects that have been running without any problem.

Does anybody have any clue where to look for the solution?
Is this a problem in Datastage, on the unix-server or on Oracle?

JeroenDmt wrote:This problem occurs in every project though, including projects that have been running without any problem.

What has changed? "Nothing" is not the correct answer.

ray.wurlod wrote:

JeroenDmt wrote:This problem occurs in every project though, including projects that have been running without any problem.

What has changed? "Nothing" is not the correct answer. ...

I wish I knew.. In the Datastage projects nothing.
We haven't been able to find anyone in Operations who are running the servers who knew about any changes on the server or in the database, but I guess there must be some kind of change or patch or something that is causing this problem.
Problem is we don't know where is the most likely place to look for problems, so that doesn't make finding the right person to push for answers any easier.

We found out what changed.. someone added several new NLS maps causing the server to need too much memory. After removing the new NLS maps and restarting the server everything worked just fine again.

You can adjust the memory available with the shared memory offset parameters in uvconfig. However, you still can't load more maps than your maximum shared memory segment size (usually 2MB, configured by shmmax kernel parameter in UNIX) allows.