PH63178: A LOAD TASK RUNNING IN PARALLEL WITH PREVIOUSLY STARTED EXPENSIVE QUERIES CAN BE HANGING (MNT LVL 7.5.12.1 OR LATER)

APAR status

• Closed as documentation error.

Error description

• For Accelerator maintenance level 7.5.12.x and 7.5.13,
  the following issue has been reported:
  An accelerator load task processing a partition of a
  table can be queued for various reasons:
  - There could be ongoing processing issues in the backend
    database which continues not to be ready to proceed with
    the load request.
  - The load task is assigned to job group although this job group
    is in status "Cancellation Requested".
  
  Such a queued load task usually will get canceled automatically
  by the accelerator server after about 40 minutes of waiting
  without processing.
  Due to processing (logic) failure, the load task neither is
  marked to have failed nor does it get started and therefore does
  not reach a successful ending state. Thus the internal
  cancellation won't be successful and leave the task hanging in
  the accelerator server.
  
  The recommended approach to deal with this hanging load task is
  to cancel it using:
  - Stored Procedure SYSPROC.ACCEL_CONTROL_ACCELERATOR
    <cancelTasks> (in Analytics Accelerator Studio)
  - Cancel of the thread in Db2 for z/OS that belongs to the
    accelerator load job.
  
  If the cancellation is not possible or not succesful, a restart
  of the accelerator will be required.
  
  Additional keywords:
  TS016625479 TS017583540
  LOAD HANG
  DT396358
  GH/.../Customer-Cases/issues/758
  GH/.../Customer-Cases/issues/724
  GH/.../IDAA-Server/issues/6077
  

Local fix

• Workaround:
  

Problem summary

• Problem Summary:
  When the Stored Procedure ACCELE_LOAD_TABLES finishes the
  processing of a partition, it emits a
  "PartitionProcessingCompletedEvent" and marks the job group
  assigned to that partition as free without waiting for the
  completion of all jobs of that job group. Because of this, the
  Stored Procedure may assign subsequent requests to process a
  partition to a job group that has not completed and is in state
  "CancellationRequested". With that, in turn, on the accelerator,
  the corresponding "PartitionProcessingStartedEvent" may be
  processed by the assigned job group while it is still in state
  "CancellationRequested". This logic failure leads to the hanging
  of the load task.
  
  
  Users Affected:
  User of Accelerator maintenance levels 7.5.12.x and 7.5.13
  running (partition) load requests.
  
  Problem Scenario:
  See APAR Error description.
  
  Problem Symptoms:
  See APAR Error description.
  

Problem conclusion

• The issue has been fixed with Accelerator maintenance level
  7.5.13.1.
  Upgrade your accelerator environment(s) accordingly.
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

Fix information

Applicable component levels