PI90759: IIB MQINPUT NODE RETAINS LARGE BUFFERS LEADING TO MEMORY SHORTAGES

APAR status

• Closed as program error.

Error description

• IIB v10 changed from using the MQGET API to the MQCTL-MQCB API.
  These behave very differently with respect to the management of
  the message buffer.
  Using the MQGET API, IIB allocated the message buffer storage
  and also monitored the size of the buffer, resizing it  if it
  was no longer required to be very large.
  Using the MQCTL-MQCB API, MQ allocates the message buffer and
  does not provide a means for the application to manage this. The
  current MQ design for the MQCTL-MQCB API results in the
  following changes in behaviour that affect the storage usage for
  IIB message flows processing MQ input messages:
  
  1. MQ allocates the message buffer. On z/OS this is in 31-bit
  storage.
  2. MQ only resizes the buffer up, to the maximum message size
  that the thread has processed, but it does not resize the buffer
  down if only smaller messages are subsequently processed.
  3. The message buffers remain allocated for the life of the
  message flow thread.
  
  As a result of these changes, an integration server (IS) that
  regularly processes very large MQ input messages on multiple
  message flow threads, might experience increased or excessive
  storage usage for the IS process.
  
  On z/OS this can be a particular problem because the MQ buffers
  are allocated in 31-bit storage which is considerably more
  limited than 64-bit storage. This can lead to the IS logging the
  following messages and an 878-10 abend loop:
  
  +BIP2628W (Msg 1/2) MQ20BRK EG3 76 EXCEPTION CONDITION
   DETECTED ON INPUT NODE 'MessageFlow.MQInput'.
  +BIP2680E (Msg 2/2) MQ20BRK EG3 76 FAILED TO SETUP A MQ
   CONTROL CALLBACK FOR COMPONENT
   'MessageFlow.MQ Input' TO QUEUE 'INPUT' ON QUEUE
   MANAGER 'MQ20': MQCC=2; MQRC=2195.
  
  IEA794I SVC DUMP HAS CAPTURED:
   DUMPID=007 REQUESTED BY JOB (MQ20BRK )
   DUMP TITLE=MQ20,ABN=878-00000010,U=WMQI20,C=W9700.800.
   MMC -CSQMALCH,M=CSQGFRCV,PSW=070C1000815F4CEE,ASID=00A7
  
  CSQY291E CSQWDSDM SDUMPX FAILED,
  RC=00003E08,MQ20,ABN=878-00000010,
   PSW=070C1000815F4CEE,ASID=00A7
  

Local fix

• N/A
  

Problem summary

• ****************************************************************
  USERS AFFECTED:
  All users of IBM Integration Bus version 10 using MQInput nodes.
  
  
  Platforms affected:
  z/OS, MultiPlatform
  
  ****************************************************************
  PROBLEM DESCRIPTION:
  In version 10, IBM Integration Bus changed from using the MQGET
  API to the MQCTL-MQCB API.
  
  These behave very differently with respect to the management of
  the message buffer.
  
  When using the MQGET API, IBM Integration Bus allocated the
  message buffer storage and also monitored the size of the
  buffer, resizing it if it was no longer required to  be very
  large. In contrast, when using the MQCTL-MQCB API, MQ allocates
  the message buffer and does not provide a means for the
  application to manage this. The current MQ design for the
  MQCTL-MQCB API results in the following changes in behaviour
  that affect the storage usage for IBM Integration Bus message
  flows processing MQ input messages:
  
  1. MQ allocates the message buffer. On z/OS this is in 31-bit
  storage.
  2. MQ only resizes the buffer upwards to the maximum message
  size that the thread has processed, but it does not resize the
  buffer downwards if only smaller messages are subsequently
  processed.
  3. The message buffers remain allocated for the life of the
  message flow thread.
  
  As a result of these changes, an integration server (IS) that
  frequently processes very large MQ input messages on multiple
  message flow threads, might experience increased or excessive
  storage usage for the IS process.
  
  On z/OS, this can be a particular problem because the MQ buffers
  are allocated in 31-bit storage, which is considerably more
  limited than 64-bit storage. This can lead to the IS logging the
  following messages and an 878-10 abend loop:
  
  +BIP2628W (Msg 1/2) MQ20BRK EG3 76 EXCEPTION CONDITION
                        DETECTED ON INPUT NODE
  'MessageFlow.MQInput'.
  +BIP2680E (Msg 2/2) MQ20BRK EG3 76 FAILED TO SETUP A MQ
                      CONTROL CALLBACK FOR COMPONENT
                       'MessageFlow.MQ Input' TO QUEUE 'INPUT' ON
                       QUEUE MANAGER 'MQ20': MQCC=2; MQRC=2195.
  
  IEA794I SVC DUMP HAS CAPTURED:
  DUMPID=007 REQUESTED BY JOB (MQ20BRK )
  DUMP TITLE=MQ20,ABN=878-00000010,U=WMQI20,C=W9700.800.
  MMC -CSQMALCH,M=CSQGFRCV,PSW=070C1000815F4CEE,ASID=00A7
  
  CSQY291E CSQWDSDM SDUMPX FAILED,
  RC=00003E08,MQ20,ABN=878-00000010,
  PSW=070C1000815F4CEE,ASID=00A7
  

Problem conclusion

• IBM Integration Bus has been modified to:
  * Resize the buffer on any active threads if it is large and has
  not been used for a period of time
  * On z/OS, disconnect from the Queue Manager and so release the
  buffer for additional instance threads that have idled out. This
  already occurred for non XA users.
  
  On z/OS MQ APAR PI90772 is required in conjunction with IBM
  Integration Bus APAR PI90759 for this APAR PI90759 to be
  effective.
  
  ---------------------------------------------------------------
  The fix is targeted for delivery in the following PTFs:
  
  Version    Maintenance Level
  v10.0      10.0.0.12
  
  The latest available maintenance can be obtained from:
  http://www-01.ibm.com/support/docview.wss?rs=849&uid=swg27006041
  
  If the maintenance level is not yet available,information on
  its planned availability can be found on:
  http://www-1.ibm.com/support/docview.wss?rs=849&uid=swg27006308
  ---------------------------------------------------------------
  

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

  IIB Z/OS

• Fixed component ID

  5655AB100

Applicable component levels