JR56417: Parallel jobs on Windows fail with APT_IOPort error 10054 when job is using TCP socket communication.

Download Data Integration Rollup 1 for InfoSphere Information Server 11.5.0.1

APAR status

• Closed as program error.

Error description

• On Windows Parallel Jobs fail with error
  IIS-DSEE-TFOR-0001
  Fatal Error: APT_IOPort::readBlkVirt: read for block header,
  partition 0, [fd 5], returned -1 with errno 10,054 (Unknown
  Error)
  
  The failure occurs on stages like Join/Merge or any processing
  stage which take 2 or more inputs where:
  - The downsteam stage is waiting on Input 1 for processing.
  - The upstream stage on Input 2 completes sending  all data.
  After processing the current buffer on Input 1,  the downstream
  stage scans the other sockets and sees  the socket for Input 2
  has been closed and throws the  error.   The socket with the
  completed data is not kept open long enough for the other side
  to read it.
  
  The failure is most likely to occur when the number of records
  handled by the two inputs is very different.
  This failure was introduced into the parallel engine in IS 9.1
  and does not affect older releasese.
  

Local fix

• This failure only happens when the parallel engine uses TCP
  sockets afor interprocess communication.  Users can switch to
  shared memory interprocess communication by defining
  environment variable:
   APT_NO_IOCOMM_OPTIMIZATION
  
  Unfortunately shared memory communication is much slower than
  socket communication.
  

Problem summary

• ERROR DESCRIPTION:
  On Windows Parallel Jobs fail with error
  IIS-DSEE-TFOR-0001
  Fatal Error: APT_IOPort::readBlkVirt: read for block header,
  partition 0, [fd 5], returned -1 with errno 10,054 (Unknown
  Error)
  The failure occurs on stages like Join/Merge or any processing
  stage which take 2 or more inputs where:
  - The downsteam stage is waiting on Input 1 for processing.
  - The upstream stage on Input 2 completes sending  all data.
  After processing the current buffer on Input 1,  the downstream
  stage scans the other sockets and sees  the socket for Input 2
  has been closed and throws the  error.   The socket with the
  completed data is not kept open long enough for the other side
  to read it.
  The failure is most likely to occur when the number of records
  handled by the two inputs is very different.
  This failure was introduced into the parallel engine in IS 9.1
  and does not affect older releasese.
  LOCAL FIX:
  This failure only happens when the parallel engine uses TCP
  sockets afor interprocess communication.  Users can switch to
  shared memory interprocess communication by defining
  environment variable:
   APT_NO_IOCOMM_OPTIMIZATION
  Unfortunately shared memory communication is much slower than
  socket communication.
  

Problem conclusion

• Patches are available for this APAR. These patches fix the
  problem by keeping the internal communication socket open long
  enough to allow jobs to shut down properly. This fix is enabled
  by setting  the environment variable APT_DEFER_OUTCUR_FD_CLOSE .
  As of July 19,2017 a revised patch for Information Serve 11.5
  makes this variable unnecessary - the sockets behavior is fixed
  by default. The fixed behavior will be the default for future
  versions. The variable APT_DEFER_OUTCUR_FD_CLOSE is deprecated.
  Setting APT_DEFER_OUTCUR_FD_CLOSE should have no effect.
  The old behavior (not keeping the sockets upen for a enough
  time) can be restored by setting the new environment variable
  APT_OUTCUR_FD_CLOSE. We do not recommend setting this variable.
  

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

• Fixed component name

  INFO SRVR PLATF

• Fixed component ID

  5724Q3612

Applicable component levels

• R912 PSY

     UP

• RB30 PSY

     UP

• RB31 PSY

     UP

• RB50 PSY

     UP