IT26789: MQ Classes for Java memory leak when communicating to the queue manager over an unreliable TCP/IP network

APAR status

• Closed as program error.

Error description

• A MQ Classes for Java application is being used to communicate
  with a queue manager.  The TCP/IP network connection to the
  queue manager is unreliable, resulting in the:
  
    com.ibm.mq.MQQueueManager
  
  object instance being regularly disconnected from the queue
  manager, after which the application creates a new object to
  replace the disconnected object instance.
  
  
  After a period of time, a Java heapdump is collected and
  analysed, and is found to contain numerous references to the
  object:
  
  com.ibm.mq.jmqi.remote.impl.RemoteSession
  
  
  The number of these objects seen in the heapdump is similar to
  the number of times the "com.ibm.mq.MQQueueManager" was
  disconnected to the queue manager.
  
  Regardless of the amount of time the JVM is left running for,
  these objects are not cleared up.
  

Local fix

Problem summary

• ****************************************************************
  USERS AFFECTED:
  This issue affects use of the IBM MQ classes for Java and JMS,
  who are connecting to the queue manager using the "CLIENT"
  transport (that is to say connecting over a network), and
  experiencing frequent network disconnection events between the
  application and the queue manager.
  
  
  Platforms affected:
  MultiPlatform
  
  ****************************************************************
  PROBLEM DESCRIPTION:
  A "java.util.concurrent.ConcurrentHashMap" is used by the MQ
  classes for Java/JMS, to track all active connections between
  the application and the queue manager.
  
  It was intended that these connections would be removed from the
  "java.util.concurrent.ConcurrentHashMap" after the MQ API flow:
  
    MQDISC
  
  was flowed to the queue manager.
  
  However - if the TCP/IP connection had been disconnected between
  the JVM and the queue manager host system, then internally
  within the MQ classes for Java/JMS, this MQDISC flow would throw
  an exception to report the broken connection.
  
  When this happened, the code to remove the entry from the
  HashMap was skipped, which meant that a reference to several
  objects associated with the connection was left in memory -
  preventing the garbage collector from removing them.
  
  While the amount of memory used by these objects was relatively
  small, if using a very unreliable network, then over time this
  could build up to consume a significant amount of the JVM's
  heap, which in turn could ultimately lead to a
  "java.lang.OutOfMemoryError" exception being thrown.
  
  The main Java objects observed in a Java heap dump associated
  with this issue were:
  
    com.ibm.mq.jmqi.remote.impl.RemoteSession
    com.ibm.mq.jmqi.remote.api.RemoteHconn
  

Problem conclusion

• The MQ classes for Java/JMS have been updated to ensure that the
  entries in the "java.util.concurrent.ConcurrentHashMap" are
  removed when the MQDISC flow is attempted, even if it should
  fail due to a disconnected connection.
  
  ---------------------------------------------------------------
  The fix is targeted for delivery in the following PTFs:
  
  Version    Maintenance Level
  v9.0 LTS   9.0.0.6
  v9.1 CD    9.1.2
  v9.1 LTS   9.1.0.2
  
  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

  IBM MQ BASE M/P

• Fixed component ID

  5724H7261

Applicable component levels