IC85614: MQ .NET XA APPLICATION IN MANAGED MODE SHOWS MEMORY LEAKS AND SPAWNS NUMEROUS AMQZLAA0 PROCESSES

Fix Pack 7.1.0.3 for WebSphere MQ V7.1
WebSphere MQ V7.5 Fix Pack 7.5.0.3

APAR status

• Closed as program error.

Error description

• WebSphere MQ .NET extended transaction (XA) application
  generates numerous amqzlaa0 processes and shows memory leaks.
  Application is running in MANAGED client mode.
  
  While application is running:
  - Memory usage keeps increasing and it does not stabilize.
  - CPU usage is between 90% and 100%.
  - Hundreds of amqzlaa0 processes are created.
  - Message throughput is very slow.
  
  Eventually the application receives MQRC_CONNECTION_BROKEN
  and MQRC_CONNECTION_ERROR errors. Communication are reported
  in the queue manager error logs. FDC is generated for the
  application.
  
  AMQ9208: Error on receive from host (xxxxxxxx)
  An error occurred receiving data from xxxxxxxx over TCP/IP
  The return code from the TCP/IP call was 10054 (X'2547')
  
  FDC shows the following:
  Probe Id :- DN<99001
  Component :- NmqiXAResourceManager.MQRMIXACommit(Bool
  Major Errorcode :- rrcE_COMMUNICATIONS_ERROR
  Probe Description :- AMQ9213: A communications error for recv
  occurred
  Comment1 :- xa_commit call has failed with return code = -4
  
  MQM Function Stack
  Unknown Function - 073 252
  Unknown Function - 073 275
  xcsFFST
  

Local fix

• Local Fix￘
  none
  

Problem summary

• ****************************************************************
  USERS AFFECTED:
  Users using WebSphere MQ .NET managed XA client application.
  This problem is applicable to WebSphere MQ v7.1 and v7.5.
  
  Platforms affected:
  Windows
  
  ****************************************************************
  PROBLEM SUMMARY:
  The below issues were seen when MQ .NET managed transactional
  clients are run.
  1) The xa_commit call fails with invalid transaction error -4
  and a FDC with the below function stack is created
  
  MQM Function Stack
  MQDTCCallbackObserver.Commit(Enlistment)
  NmqiXAResourceManager.MQRMIXACommit(Boolean)
  ManagedCommonServices.FFST
  (String,string,string,uint,uint,MQULONG,MQULONG,ushort)
  
  | Comment1          :- xa_commit call has failed with return
  code = -4        |
  
  In a  2-phase commit, if xa_prepare completes with XA_RDONLY
  then a further commit need not be issued. This return code
  implies that the transaction is complete.
  
  2) The number of running agent processes keeps increasing
  constantly. The queue manager trace shows that most of the
  agent processes were idle. The agent threads created for
  servicing the MQI request of a connection are ended or reused
  after the application disconnects. However, this thread is
  neither reused nor freed if it is part of a active transaction.
  This was happening since the MQ .NET layer did not correctly
  end the transaction even though a xa_close was issued. Hence,
  when one agent process reaches its thread limit a new agent
  process gets created.
  
  3) Memory leak was seen since the TSH buffer was not released
  during disconnect.
  

Problem conclusion

• The WebSphere MQ .NET code has been modified to fix the issues
  seen when transactional clients are run.
  
  ---------------------------------------------------------------
  The fix is targeted for delivery in the following:
  
                     v7.1
  Platform           Fix Pack 7.1.0.3
  --------           --------------------
  Windows            7.1.0.3
  
  Version            v7.5
  --------           --------------------
  Fix available in:  7.5.0.3
  
  The latest available maintenance can be obtained from
  'WebSphere MQ Recommended Fixes'
  http://www-1.ibm.com/support/docview.wss?rs=171&uid=swg27006037
  
  If the maintenance level is not yet available information on
  its planned availability can be found in 'WebSphere MQ
  Planned Maintenance Release Dates'
  http://www-1.ibm.com/support/docview.wss?rs=171&uid=swg27006309
  ---------------------------------------------------------------
  

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

  WMQ WINDOWS V7

• Fixed component ID

  5724H7220

Applicable component levels

• R710 PSY

     UP