| Author |
Message
|
| rparti |
 Posted: Tue Jun 26, 2007 4:11 pm Post subject: Clustering and Disaster Recovery |
 |
|
Voyager
Joined: 05 Oct 2005
Posts: 86
Location: U.S.A.
|
This post is to gather opinion about an implementation I was thinking
QM1 and QM2 are queue managers part of a cluster in production
QMPT is pass through queue manager connected with QM1
QMDR the disaster revovery queue manager linked with QM1 in a HACMP mutual takeover configuration. It is also linked with QMPT for passthrough.
QMPT-----QM1 QM2
|
|______QMDR
When QM1 box fails, QMDR takes over using floating IP of QM1.
Question:
Should QMDR be part of the cluster that QM1 and QM2 are part off? |
|
| Back to top |
|
 |
| jefflowrey |
| Posted: Tue Jun 26, 2007 4:39 pm Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
Um.
If the HACMP is done right, "QMDR" is the same as "QM1".
It doesn't have a different name, it doesn't have different files, it doesn't have different queues or logs or anything else.
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| rparti |
| Posted: Tue Jun 26, 2007 4:45 pm Post subject: |
|
|
Voyager
Joined: 05 Oct 2005
Posts: 86
Location: U.S.A.
|
| jefflowrey wrote: |
Um.
If the HACMP is done right, "QMDR" is the same as "QM1".
|
I believe that the above is true in a Standby configuration of HACMP. Please correct me if I am wrong. |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Tue Jun 26, 2007 6:16 pm Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
What advantages do you expect to get from "mutual takeover" that you don't have with "standby"?
What has lead you to think about this design?
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| rparti |
| Posted: Wed Jun 27, 2007 6:58 am Post subject: |
|
|
Voyager
Joined: 05 Oct 2005
Posts: 86
Location: U.S.A.
|
Per defination:
In Standby mode one or more queue managers normally run on one node. The other node is only used for failures and does not normally run any queue managers.
In Mutual Takeover one or more queue managers normally run on each node. In a failure, one node is called upon to run it’s own queue managers in conjunction with those of the other node.
I was planning to use Mutual Takeover because I thought that I will need more than one node to implement my design. Anyway on further study I found that Standby mode can suffice my needs.
However the question now is, has anybody implemented such a configuration before. Anything I should be aware of? |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Wed Jun 27, 2007 7:20 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
There's plenty of reasons to run Mutual Takeover. The important thing to note is that it's TWO queue managers. Mutual Takeover can also be thought of as "paired standby" - where each queue manager is in an active-passive setup between two nodes, but the primary node of one is the secondary node of the other.
For example, I may choose to run two instances of Message Broker simultaneously, and put the two qmgrs in a mutual takeover situations - because I want to only build two "machines" (lpars, etc.) rather than 4.
But mutual takeover is not for Disaster Recovery. Disaster Recovery implies that you are recreating "the same thing" in a different location.
You've basically built (or at least described) a standby configuration built using mutual takeover...
It's a very bad idea to have one qmgr take over the same IP address and port number of another queue manager. Channels won't work, clients will be confused, and other things may go all higgeldy-piggeldy.
But again, under standby, the secondary node runs "the same" queue manager.
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| rparti |
| Posted: Wed Jun 27, 2007 7:57 am Post subject: |
|
|
Voyager
Joined: 05 Oct 2005
Posts: 86
Location: U.S.A.
|
| jefflowrey wrote: |
But again, under standby, the secondary node runs "the same" queue manager. |
Yes....in standby the two queue managers share the same data and logs....hence we won't have to worry about channels not working, clients being confused and other problems
What is the best way then to implement disaster recovery. Suppose I want to house one queue manager in Chicago and the standby node in New York. Where should I place the queue manager logs and data.
Can the logs and data be in:
1) any other server in either Chicago, NewYork or elsewhere
2) Would it have to be on a server in which the queue manager exists in Chicago or New York
3) Can it be on any other server in Chicago or New York |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Wed Jun 27, 2007 8:00 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
It needs to be placed so that it is available to both servers, in a way that does not unduly impede performance.
Or you could look at the new options in v6 for a "backup" qmgr, instead of using HACMP.
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| rparti |
| Posted: Wed Jun 27, 2007 8:08 am Post subject: |
|
|
Voyager
Joined: 05 Oct 2005
Posts: 86
Location: U.S.A.
|
| I did not know about the backup queue manager. Thanks for sharing |
|
| Back to top |
|
|
|
|
