XML job fails for more records

suja.somu · Post by **suja.somu** » Wed Mar 12, 2014 1:32 pm

XML job has multiple source links with HJOIN steps being performed on each source links. The job ran fine for 2000 records in XML. The XML schema is complex with over 250 elements and multiple complex groups which are unbounded.

The job fails for more than 2000 records in target with the below errors.

XML_22,0: 2014-03-07 14:58:37,790 Error [com.ibm.e2.applications.runner.E2AssemblyRunner] [] java.util.concurrent.RejectedExecutionException

XML_22,0: Failure during execution of operator logic.

node_node2: Player 15 terminated unexpectedly.

buffer(4),0: Error in writeBlock - could not write 130537

Perf-tuning steps:

Changed the HJOIN to perform in-memory join.
Changed the HEAP size and STACK size in the job

Please guide me steps to solve this issue.

eostic · Post by **eostic** » Wed Mar 12, 2014 8:08 pm

If the Stage is otherwise working perfectly, but having just volume issues, I would contact your support provider. That sounds like one that should be reviewed...

Ernie

ray.wurlod · Post by **ray.wurlod** » Wed Mar 12, 2014 8:38 pm

Although it seems unlikely for just over 2000 you may not have enough memory to perform all your HJOINS in memory. What happens if you disable this setting?

suja.somu · Post by **suja.somu** » Thu Mar 13, 2014 4:22 pm

Thanks for the reply Ray.

There are only 2 options for Hjoin stage join types:
1. Disk Based
2. In-memory.

Earlier I had the DISK BASED join, job did abort frequently. IBM had suggested to use IN-MEMORY join in all documentation , so I changed to In-Memory and this could help to process only up to 2000 records as output.

I monitored the CPU usage and memory usage while the job ran, both didn't go to 100%. it was quiet normal usage.

I tried to enable the APT_DUMP_SCORE , but it does not give any detailed step in the job log.

Could not trace where the issue is. Please help.

chulett · Post by **chulett** » Thu Mar 13, 2014 5:13 pm

As Ernie noted, this should be going back to your support provider.