| Author |
Message
|
| sebastia |
 Posted: Tue Feb 20, 2007 4:54 am Post subject: Cluster queue only visible after using it |
 |
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
We have a cluster of 3 Qmgrs
2 are full repository, named QMS1 and QMS2.
1 is partial repository, called QMS3
If I create a queue (call it Q1) and share it to the cluster on QMS3,
instantly QMS1 and QMS2 do see it (using RUNMQSC or Explorer).
But if I define a queue in QMS1 or QMS2 [FR],
it is not visible at QMS3 [PR]
- using DISPLAY QCLUSTER(Q*) -
but we can use it using AMQSPUT
and after the first AMQSPUT it becomes visible
- using the same "DISPLAY QCLUSTER(Q*)"
Is this normal ?
That is not very NICE to the customer,
when we talk about administration.
Is there any timer I can decrease so this new queue existence propagates faster ?
Thanks. |
|
| Back to top |
|
 |
| Vitor |
| Posted: Tue Feb 20, 2007 5:16 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
This is perfectly normal behaviour.
Cluster queues only appear on partial repositories on 1st use. Queues only propograte to full repositories, and there's no timer on that. It'll happen as soon as resources permit (normally near enough immediately).
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| Mr Butcher |
| Posted: Tue Feb 20, 2007 5:17 am Post subject: |
|
|
Padawan
Joined: 23 May 2005
Posts: 1716
|
YES. Thats the way a PR works.
New cluster objects are not propagated to a PR. Instead - if a Pr wants to use an unknown object the FR is asked if that object is known in the cluster. If so, the PR is given information about the object.
There is no timer for propagation.
why is this not nice to the costomer?
_________________
Regards, Butcher |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Feb 20, 2007 5:24 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
Well - if it works as designed, I will stop searching for the timer ( )
But, mr Butcher, you'd agree that if we have a cluster of 20 or 30 queue managers (only 2 of them have FR), then the MQ Explorer view is quite false until the queue(s) is/are used.
What is the correct way for an application running on a queue manager that has partial repository to know for the existence of a cluster queue ?
I can try to MQput() into it, but a failure wont diferentiate between "non existing" or "not reachable" ...
I mean, your sentence talks about "asking" the queue managers with FR about this info .. how is it done ? I mean, from RUNMQSC ? from MQ Explorer ? from an application ?
Thanks. |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Tue Feb 20, 2007 5:33 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
An application doesn't have to "query" or "ask" for the qcluster.
It has to PUT to it.
If the PUT fails, it has to handle the reason code that comes back.
Applications shouldn't "discover" the queues that they need to use, as they'd have to guess! They should always be TOLD what they are - either in configuration, or a routing table, or the ReplyToQ.
The MQ Explorer view is not "false". It's "accurate" - it shows you want the qmgr knows about at the time - which except in cases of an FR will not ever been the full cluster.
There's a Clusters view, that may actually show the information that you want to see on every queue manager.
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Feb 20, 2007 5:41 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| sebastia wrote: |
| But, mr Butcher, you'd agree that if we have a cluster of 20 or 30 queue managers (only 2 of them have FR), then the MQ Explorer view is quite false until the queue(s) is/are used. |
Why? An administrator would use an FR for reference
| sebastia wrote: |
| What is the correct way for an application running on a queue manager that has partial repository to know for the existence of a cluster queue ? |
Proper design. If an application is trying to use an non-existant cluster queue it's got the wrong name coded into it. Or the cluster is set up wrong.
| sebastia wrote: |
I mean, your sentence talks about "asking" the queue managers with FR about this info .. how is it done ? I mean, from RUNMQSC ? from MQ Explorer ? from an application ?
|
By the queue manager, the 1st time an application asks to open the queue. If the queue doesn't exist and the queue manager is clustered, it'll request the information from it's FR.
All this is laid out in the clustering manual.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| Mr Butcher |
| Posted: Tue Feb 20, 2007 7:03 am Post subject: |
|
|
Padawan
Joined: 23 May 2005
Posts: 1716
|
| Quote: |
| But, mr Butcher, you'd agree that if we have a cluster of 20 or 30 queue managers (only 2 of them have FR), then the MQ Explorer view is quite false until the queue(s) is/are used. |
no i do not agree. the mq cluster works "as is", and there is nothing wrong with the explorer view.
| Quote: |
| What is the correct way for an application running on a queue manager that has partial repository to know for the existence of a cluster queue ? |
why is the application interested in that? do the open, and you know.
in addition, do you get applications but you dont know if the proper queues for the application exist? why did you build a cluster? for the application or for yourself?
if the cluster was build to support the application, why do you worry about the queues being existent? they should be there, thats the setup.
i do not understand your point very well. from my point of view, you look at the cluster from the wrong direction.
what are you trying to archive with your "does my cluster queue exist" - investigation?
_________________
Regards, Butcher |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Feb 20, 2007 7:21 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
mr Butcher - you have written a very clever question :
>>> what are you trying to achieve ? <<<
I am sitting with the customer in from of MQ Explorer
and telling him the great advantadges of administering a cluster.
I say "You dont have to create a REMOTE QUEUE for every queue
you want to write into. All you have to do is ...
DEFINE QLOCAL(INVENTQ) CLUSTER(INVENTORY)"
So, I open a RUNMQSC dos window, type in the DEFINE command
and ... it is displayed in the FR qmgr resources but not in the PR qmgr.
And I have to explain that, and I can explain it will be there once it is used, ... but it is inconsistent
Thanks for all the opinions. S. |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Tue Feb 20, 2007 7:27 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
| jefflowrey wrote: |
| There's a Clusters view, that may actually show the information that you want to see on every queue manager. |
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Feb 20, 2007 7:37 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
Yes, Jeff - you are right - it is there.
But if the queue IS accessible from any queue manager
of the cluster,
why it does not show
if I run DISPLAY QCLUSTER on a remote (PR) q mgr ?
It would be ok if it were NOT accessible, you see what I mean ?
It is there ... (ok)
I can write into it ... (ok)
I do not SEE it ... (??)
but I will see it after I use it .. (!?!?!?)
Sebastian. |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Feb 20, 2007 7:40 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| sebastia wrote: |
Yes, Jeff - you are right - it is there.
But if the queue IS accessible from any queue manager
of the cluster,
why it does not show
if I run DISPLAY QCLUSTER on a remote (PR) q mgr ?
|
Because until it's been used, it doesn't exist on the PR q manager. The DISPLAY command does not show all the accessable queues, it shows the existing ones.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| Mr Butcher |
| Posted: Tue Feb 20, 2007 7:46 am Post subject: |
|
|
Padawan
Joined: 23 May 2005
Posts: 1716
|
| Quote: |
| And I have to explain that, and I can explain it will be there once it is used, ... but it is inconsistent |
well.... no. it is not 
just think of a FR as a manager of information. the FR has to know about everything in the cluster. so everything is visible here, and if you have more then one FR they talk to each other so they are at the same knowledge level.
now, there are many PRs. why tell everybody about every clusterobject that it is not interested in? you could do so, of course, but think of big clusters with 100reds of PRs, what an overhead. so for the PR they only get information if they ask for it.
thats the way it is designed and how it works, and it works well this way.
just change your point of view a little bit. its not inconsistent. its a design question.
_________________
Regards, Butcher |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Feb 20, 2007 8:12 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
Thanks a lot, mr Butcher for your time and patience.
To align my point of view with the MQ cluster's designers,
I will keep in mind that ... ONLY FR has complete view of the cluster,
so it is the right point where to look for Administration.
Yes, that is consistent.
Cheers ! |
|
| Back to top |
|
|
|
|
