Oracle Stage succeeds but do not execute all sql request

mrekik · Post by **mrekik** » Wed Feb 06, 2008 9:08 am

Hello everybody
I have a strange behavior (not reproducible) of Oracle stage : The same stage has 10 output links (10 different sql requests), during its execution, the job executes some links and avoids (without any reason) the others with no trace in the director.
The output of all links is stored in hashed files, the non-executed links didn't generate any hashed file ...

The status of job is success in director.

Thank for any help

ArndW · Post by **ArndW** » Wed Feb 06, 2008 9:15 am

There isn't enough information for an answer in that post, unfortunately.

I am unclear on how you see that a link isn't executed - is this via the job log or because hashed files don't get created?

mrekik · Post by **mrekik** » Wed Feb 06, 2008 9:41 am

These is no explanation trace in the director (that's why I sent a post here:)).

mrekik · Post by **mrekik** » Wed Feb 06, 2008 9:43 am

These is no explanation trace in the director (that's why I sent a post here:)).

mrekik · Post by **mrekik** » Wed Feb 06, 2008 9:46 am

I verify that the hash file was not generated by checking it on the hash file directory

kcbland · Post by **kcbland** » Wed Feb 06, 2008 9:51 am

I have seen something similar over the years with all versions of Server. If the job has failed previously there may be active threads (DSD.StageRun processes) still connected to the database. Running the job again seems to "adopt" these threads and close them down without initiating its own threads. The job appears to function correctly, but the threads never actually run.

Check for active threads with "ps -ef|grep phantom" and look for your jobname. Kill everything and run again.

mrekik · Post by **mrekik** » Wed Feb 06, 2008 10:01 am

Thank you for your feed back.

All functional flows are scheduled during the night with automatic reset in case of failure.

Is there an automatic (under windows server) solution to avoid this strange behavior.

kcbland · Post by **kcbland** » Wed Feb 06, 2008 10:24 am

We're not sure exactly the nature of the problem. In the event of a job failure an automatic reset does not necessarily cleanup hung threads. To fully automate and guarantee a job is ready to run again, your processes (automated or manual) must guarantee a cleanup. In Windoze environments you're going to be challenged to script such a solution.

mrekik · Post by **mrekik** » Wed Feb 06, 2008 11:05 am

OK, Thank you for your help

ray.wurlod · Post by **ray.wurlod** » Wed Feb 06, 2008 3:01 pm

Get your DBA to check whether there are any reported failures at the Oracle end, for example "failed to start listener" or "no more listeners".

Do you get row counts (even if zero) reported on all ten output links?