+

Search Tips | Advanced Search

Handling message affinities

Message affinities are rarely part of good programming design. We need to remove message affinities to use clustering fully. If we cannot remove message affinities, we can force related messages to be delivered using the same channel and to the same queue manager.

If we have applications with message affinities, remove the affinities before starting to use clusters.

Removing message affinities improves the availability of applications. An application sends a batch of messages that has message affinities to a queue manager. The queue manager fails after receiving only part of the batch. The sending queue manager must wait for it to recover and process the incomplete message batch before it can send any more messages.

Removing messages affinities also improves the scalability of applications. A batch of messages with affinities can lock resources at the destination queue manager while waiting for subsequent messages. These resources might remain locked for long periods of time, preventing other applications from doing their work.

Furthermore, message affinities prevent the cluster workload management routines from making the best choice of queue manager.

To remove affinities, consider the following possibilities:

  • Carrying state information in the messages
  • Maintaining state information in nonvolatile storage accessible to any queue manager, for example in a Db2 database
  • Replicating read-only data so that it is accessible to more than one queue manager

If it is not appropriate to modify the applications to remove message affinities, there are a number of possible solutions to the problem.


Name a specific destination on the MQOPEN call

Specify the remote-queue name and the queue manager name on each MQOPEN call, and all messages put to the queue using that object handle go to the same queue manager, which might be the local queue manager. Specifying the remote-queue name and the queue manager name on each MQOPEN call has disadvantages:

  • No workload balancing is carried out. You do not take advantage of the benefits of cluster workload balancing.
  • If the target queue manager is remote and there is more than one channel to it, the messages might take different routes and the sequence of messages is still not preserved.
  • If your queue manager has a definition for a transmission queue with the same name as the destination queue manager, messages go on that transmission queue rather than on the cluster transmission queue.


Return the queue manager name in the reply-to queue manager field

Allow the queue manager that receives the first message in a batch to return its name in its response. It does this using the ReplyToQMgr field of the message descriptor. The queue manager at the sending end can then extract the reply-to queue manager name and specify it on all subsequent messages. Using the ReplyToQMgr information from the response has disadvantages:

  • The requesting queue manager must wait for a response to its first message
  • We must write additional code to find and use the ReplyToQMgr information before sending subsequent messages
  • If there is more than one route to the queue manager, the sequence of the messages might not be preserved


Set the MQOO_BIND_ON_OPEN option on the MQOPEN call

Force all your messages to be put to the same destination using the MQOO_BIND_ON_OPEN option on the MQOPEN call. Either MQOO_BIND_ON_OPEN or MQOO_BIND_ON_GROUP must be specified when using message groups with clusters to ensure that all messages in the group are processed at the same destination.

By opening a queue and specifying MQOO_BIND_ON_OPEN, you force all messages that are sent to this queue to be sent to the same instance of the queue. MQOO_BIND_ON_OPEN binds all messages to the same queue manager and also to the same route. For example, if there is an IP route and a NetBIOS route to the same destination, one of these is selected when the queue is opened and this selection is honored for all messages put to the same queue using the object handle obtained.

By specifying MQOO_BIND_ON_OPEN you force all messages to be routed to the same destination. Therefore applications with message affinities are not disrupted. If the destination is not available, the messages remain on the transmission queue until it becomes available again.

MQOO_BIND_ON_OPEN also applies when the queue manager name is specified in the object descriptor when you open a queue. There might be more than one route to the named queue manager. For example, there might be multiple network paths or another queue manager might have defined an alias. If you specify MQOO_BIND_ON_OPEN, a route is selected when the queue is opened.

Note: This is the recommended technique. However, it does not work in a multi-hop configuration in which a queue manager advertises an alias for a cluster queue. Nor does it help in situations in which applications use different queues on the same queue manager for different groups of messages.

An alternative to specifying MQOO_BIND_ON_OPEN on the MQOPEN call, is to modify your queue definitions. On your queue definitions, specify DEFBIND(OPEN), and allow the DefBind option on the MQOPEN call to default to MQOO_BIND_AS_Q_DEF.


Set the MQOO_BIND_ON_GROUP option on the MQOPEN call

Force all your messages in a group to be put to the same destination using the MQOO_BIND_ON_GROUP option on the MQOPEN call. Either MQOO_BIND_ON_OPEN or MQOO_BIND_ON_GROUP must be specified when using message groups with clusters to ensure that all messages in the group are processed at the same destination.

By opening a queue and specifying MQOO_BIND_ON_GROUP, you force all messages in a group that are sent to this queue to be sent to the same instance of the queue. MQOO_BIND_ON_GROUP binds all messages in a group to the same queue manager, and also to the same route. For example, if there is an IP route and a NetBIOS route to the same destination, one of these is selected when the queue is opened and this selection is honored for all messages in a group put to the same queue using the object handle obtained.

By specifying MQOO_BIND_ON_GROUP you force all messages in a group to be routed to the same destination. Therefore applications with message affinities are not disrupted. If the destination is not available, the messages remain on the transmission queue until it becomes available again.

MQOO_BIND_ON_GROUP also applies when the queue manager name is specified in the object descriptor when you open a queue. There might be more than one route to the named queue manager. For example, there might be multiple network paths or another queue manager might have defined an alias. If you specify MQOO_BIND_ON_GROUP, a route is selected when the queue is opened.

If MQOO_BIND_ON_GROUP is specified but the messages are not grouped, the behavior is equivalent to MQOO_BIND_NOT_FIXED.

Note: This is the recommended technique for ensuring that messages in a group are sent to the same destination. However, it does not work in a multi-hop configuration in which a queue manager advertises an alias for a cluster queue.

An alternative to specifying MQOO_BIND_ON_GROUP on the MQOPEN call, is to modify your queue definitions. On your queue definitions, specify DEFBIND(GROUP), and allow the DefBind option on the MQOPEN call to default to MQOO_BIND_AS_Q_DEF.


Write a customized cluster workload exit program

Instead of modifying the applications we can circumvent the message affinities problem by writing a cluster workload exit program. Writing a cluster workload exit program is not easy and is not a recommended solution. The program would have to be designed to recognize the affinity by inspecting the content of messages. Having recognized the affinity, the program would have to force the workload management utility to route all related messages to the same queue manager.

Parent topic: Application programming and clusters

Last updated: 2020-10-04