MQ Queue Manager Clusters

=> CLUSTERING
===========
Clustering is a way to logically group WebSphere MQ queue managers
- reduced system administration due to fewer channel, remote queue, and transmission queue definitions
- increased availability and workload balancing.

=> Basic Cluster setup
===============
> Step-1
Determine which queue manager should hold full repositories
A full repository contains a complete set of information about every queue manager and object in the cluster
You will need at least one, preferably two
> Step-2
Alter the queue manager definitions to add repository definitions
ALTER QMGR REPOS(cluster_name)
> Step-3
Define the CLUSRCVR channels
Every queue manager in the cluster needs a CLUSRCVR with a conname pointing to itself.
DEFINE CHANNEL(channel_name) CHLTYPE(CLUSRCVR) TRTYPE(TCP) CONNAME(‘my_ip_name_or_address(port)’) CLUSTER(cluster_name)
> Step-4
Define the CLUSSDR channels
Define one CLUSSDR to a full repository queue manager. The channel name must match that of the CLUSRCVR on the full repository
DO NOT define a CLUSSDR to point to a partial repository.
DEFINE CHANNEL(channel_name) CHLTYPE(CLUSSDR) TRPTYP(TCP) CONNAME(‘remote_ip_name_or_address(port)’) CLUSTER(cluster_name)
> Step-5
Define a cluster queue
DEFINE QLOCAL(qname) CLUSTER(cluster_name)
Other queue managers in the cluster can send message to it without making remote-queue definitions for it.
Only the local queue manager can read messages from an instance of the cluster queue
You can use a sample program to test putting messages to clustered queues

=> Cluster Commands
===============
DISPLAY QMGR REPOS REPOSNL QMID
AMQ8408: Display Queue Manager details.
QMNAME(QM1) QMID(QM1_2005-07-12_17.14.38) REPOS(QMCLUS)REPOSNL( )
QMID is an internally generated unique name that consists of the queue manager name plus the time the queue manager was created

DISPLAY CLUSQMGR(*) ALL
Display cluster Queue managera details

dis chstatus(*) all
Display channel status details

DISPLAY QCLUSTER(*) ALL
It displays information about clustered queues only.
A cluster queue will not be displayed on a partial repository until an application has opened it.

DISPLAY QUEUE(*) CLUSINFO
This command displays information about queues with TYPE(QCLUSTER)

=> Work load balancing
================
When a cluster contains more than one instance of the same queue, workload balancing determines the best queue manager to route a message to
- At its simplest, workload management results in a round-robin effect
MQ V6 has additional parameters that can be used to influence the results of the algorithm.
- Queues: CLWLPRTY, CLWLRANK, CLWLUSEQ
- Queue Managers: CLWLUSEQ, CLWLMRUC
- Channels: CLWLPRTY, CLWLRANK, CLWLWGHT, NETPRTY
For workload balancing to occur:
- open the queue with the MQOO_BIND_NOT_FIXED open option
- open with the default MQOO_BIND_AS_Q_DEF and with DEFBIND(NOTFIXED) set in the queue definition
- Leave MQMD.ObjectQMgrName blank to allow the queue manager to chose the queue instance
- To force the message to a specific instance of the clustered queue, specify that queue manager’s name in ObjectQmgrName

=> Namelists in clusters
================
- A queue manager may be a member of more than one cluster. List those clusters in a NAMELIST.
- You can alter a full repository QMGR to use REPOSNL(namelist) rather than REPOS.
- For channels and queues, you can specify CLUSNL(namelist) rather than specifying the CLUSTER parameter.

=> REFRESH CLUSTER
===============
- REFRESH CLUSTER removes and rebuilds locally held information about a cluster.
- REFRESH CLUSTER(clustername) REPOS(NO)
- REFRESH CLUSTER(clustername) REPOS(YES), also refreshes information about full repository queue managers and may not be issued from a full repository.
- REFRESH CLUSTER(*)

=> RESET CLUSTER
=============
RESET CLUSTER is issued from a full repository queue manager. It forcibly removes a queue manager or specific QMID from a cluster.
- RESET CLUSTER(clustername) QMNAME(qmname) ACTION(FORCEREMOVE) QUEUES(NO)
- RESET CLUSTER(clustername) QMID(qmid) ACTION(FORCEREMOVE) QUEUES(NO)

=> Troubleshooting
=============
- Is the repository manager still running?
Check the AMQERRxx.log or CHIN joblog.
- Are channels able to run in both directions?
Display CLUSQMGR and CHSTATUS information.
- Are the SYSTEM.CLUSTER.* queues enabled?
Issue DISPLAY QUEUE(SYSTEM.C*) ALL
- Are there messages on
SYSTEM.CLUSTER.COMMAND.QUEUE or
SYSTEM.CLUSTER.TRANSMIT.QUEUE?
- Are there duplicate QMIDs for a given QMGR?
Issue DISPLAY CLUSQMG(*) QMID.
- DISPLAY CLUSQMGR may show CLUSQMGR names starting with SYSTEM.TEMP.
The queue manager has not received all necessary information from the full repository.