| Author |
Message
|
| thimerion |
 Posted: Tue Nov 18, 2008 4:49 am Post subject: Load balancing - High availability |
 |
|
Acolyte
Joined: 08 May 2007
Posts: 67
Location: Belgium
|
Hi,
I have been working with mq for a while ... unclustered ...
I am totally a newbie on clustering ...
But now I have the request to make it highly available ...
So what I want to create is 2 mq servers, balancing the load over each other ... and if one mq server is down, the other continues with 100% of the load ... until the one server is back again ...
So what do I need to make this?
- A DNS name that points to 2 ip adresses and load balances over them ?
- 2 servers, having mq installed, with the Queue Managers configured as clustered ?
- All queues created on both servers ...
Is there anything not doable in this plan ?
Thanks,
Tim |
|
| Back to top |
|
 |
| Vitor |
| Posted: Tue Nov 18, 2008 4:56 am Post subject: Re: Load balancing - High availability |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| thimerion wrote: |
| Is there anything not doable in this plan ? |
The high availability part? 
As has been discussed endlessly on the forum, MQ clustering is not an HA solution and is not to be confused with an HA cluster. The solution you described will cause some messages to be delayed or lost in the event of a server failure. The question of this being an acceptable risk is one you'll need to make yourself, given your situation.
Moving on, you don't need a load balanced DNS - it's MQ, not the network, that's distributing traffic. Aside from that, you're fine, review the Clusters manual and consider buying some HA software to do HA, and leave the load balancing to MQ as the software is designed to do.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| thimerion |
| Posted: Tue Nov 18, 2008 5:11 am Post subject: |
|
|
Acolyte
Joined: 08 May 2007
Posts: 67
Location: Belgium
|
Hi Vitor,
1*/
Assumed if I take the risk to lose a few messages at the moment that one machine of my cluster goes down ...
What I do not understand is, how mq can distribute the traffic ... For example ... Someone sends a message to queue manager on ip 1.1.1.2 ... I assume that queue manager will then try to load ballance between itself and the other machine 1.1.1.3 ...
But if the machine on 1.1.1.2 goes down ... Probably all messages are rejected ... That's why I thought that I need something that balances load between available machines ...
2*/
Most probably I will not accept the risk of losing messages ... Is it possible to configure a second machine as an Mq hot standby ...
Second machine is doing nothing, first machine gets down, second machine takes over ...
Is that possible with a network storage array, or does mq not like this ...
Thanks,
Tim |
|
| Back to top |
|
|
| exerk |
| Posted: Tue Nov 18, 2008 5:59 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
Veritas/HACMP/<insert flavour of HA software here> both your queue managers.
In the event that one goes down there will be a delay while it comes up on the standby server (make your two servers active/active), the other queue manager will pick up the load while that is happening, and when the other restarts its messages will again be available.
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| thimerion |
| Posted: Tue Nov 18, 2008 6:25 am Post subject: |
|
|
Acolyte
Joined: 08 May 2007
Posts: 67
Location: Belgium
|
Hi,
What I really want to know ... if you cluster 2 queue managers, both knowing all queues, it will still work if one queue manager is down ...
This regardless of losing messages or so at the moment one server crashes ...
We just need backup ...
Tim |
|
| Back to top |
|
|
| exerk |
| Posted: Tue Nov 18, 2008 6:46 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
If you 'lose' one of the queue managers, the messages it contains will be unavailable until it restarts, and the other queue manager will get all the traffic as its instance of the queues will be the only available targets.
You have to consider:
1. The number of messages that will be on the failed queue manager (lots/negligible/etc.).
2. How long it will take to restart the queue manager? (time sensitivity of the trapped messages).
3. Can the remaining server handle the increased load for the time it takes to recover the failed queue manager?
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Nov 18, 2008 7:21 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| exerk wrote: |
| 1. The number of messages that will be on the failed queue manager (lots/negligible/etc.). |
Given that there is a period of time (discussed in this section somewhere) where messages will be routed to the downed queue manager because the cluster has not yet realised it's down.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| thimerion |
| Posted: Tue Nov 18, 2008 7:48 am Post subject: |
|
|
Acolyte
Joined: 08 May 2007
Posts: 67
Location: Belgium
|
Sounds pretty ok then ...
It's just a pass-through queue manager, meaning, the data will be send to the destination queue manager at once, and incoming data will be picked up at once ...
So we can assume that a message is no longer than a second on the queue manager.
So if the one part of the cluster automatically starts to take the full load, we can continue processing without to interact ... and we have time to fix the other server part without delaying traffic (except from the few files that might been in memory at the moment of the crash) |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Nov 18, 2008 7:54 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| thimerion wrote: |
| (except from the few files that might been in memory at the moment of the crash) |
They'll survive if the messages are persistent.
There will also be the messages addressed to the downed server, but not yet transmitted. Though in the set up you describe, it does sound like the window of oppertunity for a stuck message is small, and not serious.
Note that if the server can't be fixed, but is rebuilt with a new queue manager (even if identically named) it won't participate in the cluster & manual action will need to be taken to remove the "old" queue manager & add the new one, as per the Clusters manual.
Just offering a heads up.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| thimerion |
| Posted: Tue Nov 18, 2008 8:20 am Post subject: |
|
|
Acolyte
Joined: 08 May 2007
Posts: 67
Location: Belgium
|
Thanks I have a better insight on the possibilities now.
Regards,
Tim |
|
| Back to top |
|
|
|
|
