| Author |
Message
|
| Gideon |
 Posted: Fri Jan 11, 2013 10:45 am Post subject: Clustering |
 |
|
Chevalier
Joined: 18 Aug 2009
Posts: 403
|
From the documentation and the posts I have read, clustering is performed by the WMQ layer, not WMB.
The way that it is determined if a WMB process gfoes down is that the queue is polled for a QSI interval between the last put and the next GET, and if the GET is not performed (ie the broker is not pulling from the queue), the QSI interval can trigger that this node is no longer used
My question: Is it a normal practice simply to have 2 WMB brokers on 2 machines and each broker performs a GET on a shared queue ?
Am I missing anything ? |
|
| Back to top |
|
 |
| lancelotlinc |
| Posted: Fri Jan 11, 2013 11:09 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
|
| Back to top |
|
|
| mqjeff |
| Posted: Fri Jan 11, 2013 11:17 am Post subject: Re: Clustering |
|
|
Grand Master
Joined: 25 Jun 2008
Posts: 17447
|
| Gideon wrote: |
From the documentation and the posts I have read, clustering is performed by the WMQ layer, not WMB.
The way that it is determined if a WMB process gfoes down is that the queue is polled for a QSI interval between the last put and the next GET, and if the GET is not performed (ie the broker is not pulling from the queue), the QSI interval can trigger that this node is no longer used |
I don't have any knowledge of anything like that.
| Gideon wrote: |
My question: Is it a normal practice simply to have 2 WMB brokers on 2 machines and each broker performs a GET on a shared queue ?
Am I missing anything ? |
The only place you have anything like a shared queue is on zOS where you have an actual shared queue - where the same messages are available in two places at once.
On distributed, a multi-instance broker acts just like a normal broker. If the queue manager it's running against (which must also be a multi-instance queue manager) fails over to the backup, then the broker fails over. If the broker experiences an issue, it does it's normal restart processes to provide continued availability.
If you want to provide for a message flow to continue to process messages if the flow experiences issues, you increase the number of instances. If you want the flow to continue to process if the EG experiences issues, you deploy it to more than one EG.
If you want the flow to continue processing messages in case of total hardware failure or queue manager failure, you use a multi-instance broker to recover processing on a back up server.
If you want to share load across more than one machine, you can use an MQ cluster to share messages across the queue managers of two Brokers, each of which could indeed be multi-instance to provide recovery. |
|
| Back to top |
|
|
| Vitor |
| Posted: Fri Jan 11, 2013 11:46 am Post subject: Re: Clustering |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| Gideon wrote: |
| My question: Is it a normal practice simply to have 2 WMB brokers on 2 machines and each broker performs a GET on a shared queue ? |
No, because there's no such thing as a shared queue except on z/OS & the majority of WMB is installed on distributed platforms. Indeed, the way you talk about "2 machines" leads me to thinking this is the case here.
A shared queue is not the same as a clustered queue. A clustered queue is where 2 instances of an identically named queue exist within a single cluster and messages are distibuted between them. Applications (like WMB) can only perform a GET on the instance of the queue hosted by their queue manager.
Normal practice with broker is instead to have it properly configured within the HA software product of your choice. As the queue manager should be.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| Vitor |
| Posted: Fri Jan 11, 2013 11:47 am Post subject: Re: Clustering |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| mqjeff wrote: |
| Gideon wrote: |
From the documentation and the posts I have read, clustering is performed by the WMQ layer, not WMB.
The way that it is determined if a WMB process gfoes down is that the queue is polled for a QSI interval between the last put and the next GET, and if the GET is not performed (ie the broker is not pulling from the queue), the QSI interval can trigger that this node is no longer used |
I don't have any knowledge of anything like that. |
Nor do I and it seems a bit limited. It assumes that the broker's crashed but the queue manager has not. Better to have both queue manger and broker processes monitored by whatever's supplying this "node" concept.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| Gideon |
| Posted: Fri Jan 11, 2013 11:51 am Post subject: |
|
|
Chevalier
Joined: 18 Aug 2009
Posts: 403
|
The documentation states that to create a multi-instance broker, you perform ...
mqsicreatebroker -e <location of network storage>
mqsiaddbrokerinstance -e <location of network storage>
This appears to create an Active / Passive WMB multi-instnace broker cluster with the cluster configuration located in the shared network storage
My question: Does this make the shared network storage a single point of failure for a cluster ? |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Fri Jan 11, 2013 11:53 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
The shared network storage reflects the level of investment (ie. Active Passive).
If you want higher up time, then additional investment in Active Active is needed.
Yes, any shared resource is a single point of failure.
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| mqjeff |
| Posted: Fri Jan 11, 2013 11:53 am Post subject: |
|
|
Grand Master
Joined: 25 Jun 2008
Posts: 17447
|
| Gideon wrote: |
The documentation states that to create a multi-instance broker, you perform ...
mqsicreatebroker -e <location of network storage>
mqsiaddbrokerinstance -e <location of network storage>
This appears to create an Active / Passive WMB multi-instnace broker cluster with the cluster configuration located in the shared network storage
My question: Does this make the shared network storage a single point of failure for a cluster ? |
The answer is either obvious or really complicated.
Generally one would use a SAN or contact admin to host the 'shared storage' such that it was presumed that the storage was highly available. |
|
| Back to top |
|
|
|
|
