| Author |
Message
|
| MaheshPN |
 Posted: Thu Oct 28, 2004 7:55 am Post subject: Client Concentrator : Failover and load balancing issues |
 |
|
Master
Joined: 21 May 2003
Posts: 245
Location: Charlotte, NC
|
Hi guys,
The business wants to have failover and load balancing for the workflow servers.
We thought of using the client concentrator(client with queue manager) to resolve this.
I would guess,we have missed some step in this, due to which, if primary server goes down,
system will not respond.
Here is our setup...
We have two WF servers running on AIX
server1(primary)
Configid : FMC
group : FMCGRP
SYSTEM : FMCSYS
QUEPREFIX : FMC
QM name : O5UAT01A(First queue manager in the cluster)
CLUSTER : HOMEGRP
Also runs Scheduling server and cleanup server
On server1, I added the other systems info in the client setting menu.
SysGrp-- System-- QueueMgr Prefix--
- FMCGRP FMCSYS O5UAT01A FMC
- FMCGRP FMCSYS1 O5UAT02A FMC
server2(secondary)
Configid : FMC
group : FMCGRP
SYSTEM : FMCSYS1
QUEPREFIX : FMC
QM name : O5UAT02A(Additional queue manager in the cluster & Backup Repository)
CLUSTER : HOMEGRP
On server2, I added the other systems info in the client setting menu.
SysGrp-- System-- QueueMgr Prefix--
- FMCGRP FMCSYS1 O5UAT02A FMC
- FMCGRP FMCSYS O5UAT01A FMC
Client system1(where WAS application is running)
Configid : FMC
group : FMCGRP
SYSTEM : FMCSYS
QUEPREFIX : FMC
QM name : O5UAT01C(Additional queue manager in the cluster)
CLUSTER : HOMEGRP
info from fmcrc
MQClusterFirstQueueManager=O5UAT1A
FMLClientChannelTable=/var/fmc/chltabs/HOMECHL.TAB
FMLConnectName=FMC.FMCGRP.FMCSYS,o5UAT1C
On the Client1 side client setting menu looks like this
- FMCGRP FMCSYS O5UAT01C FMC
Clientsystem2(where WAS application is running)
Configid : FMC
group : FMCGRP
SYSTEM : FMCSYS
QUEPREFIX : FMC
QM name : O5UAT02C(Additional queue manager in the cluster)
CLUSTER : HOMEGRP
info from fmcrc
MQClusterFirstQueueManager=O5UAT1A
FMLClientChannelTable=/var/fmc/chltabs/HOMECHL.TAB
FMLConnectName=FMC.FMCGRP.FMCSYS,o5UAT2c
On the Client1 side client setting menu looks like this
- FMCGRP FMCSYS O5UAT02C FMC
All the quemanagers have their channel tab files in thier respective chltabs directory.
The problem is, Both the clients always connects to server1(tested it by running fmcautil on client sytems)
and if server1 goes down, clients will time out.
Please let me, where I went wrong.
Thanks,
-Mahesh
IBM Certified Solution Expert - MQWorkflow |
|
| Back to top |
|
 |
| vennela |
| Posted: Thu Oct 28, 2004 8:10 am Post subject: |
|
|
Jedi Knight
Joined: 11 Aug 2002
Posts: 4055
Location: Hyderabad, India
|
This will not work.
If primary goes down almost your workflow server is down. I think the clients already connected might still be able to continue the work. But you new clients will not be able to login.
Admin server runs only on the primary. That is the reason your fmcautil always goes to the primary.
For failover, you will have to reply on OS failover only (like HACMP, Veritas or MC Service Gaurd). |
|
| Back to top |
|
|
| manoj |
| Posted: Fri Oct 29, 2004 5:32 am Post subject: |
|
|
Master
Joined: 30 Jan 2002
Posts: 237
Location: Virgina
|
| Quote: |
| Admin server runs only on the primary. That is the reason your fmcautil always goes to the primary. |
I think this statement is incorrect.. Admin Server is per system
_________________
-manoj |
|
| Back to top |
|
|
| vennela |
| Posted: Fri Oct 29, 2004 6:25 am Post subject: |
|
|
Jedi Knight
Joined: 11 Aug 2002
Posts: 4055
Location: Hyderabad, India
|
I admit.
Thanks for the correction.
My previous post was dumb. |
|
| Back to top |
|
|
| vennela |
| Posted: Fri Oct 29, 2004 8:13 am Post subject: |
|
|
Jedi Knight
Joined: 11 Aug 2002
Posts: 4055
Location: Hyderabad, India
|
Mahesh:
Sorry for yesterday's post.
However, what I said is partially correct. Because your cleanup server and shceduling servers run only on primary. So your notifications and expirations will not work if the primary goes down.
| Quote: |
On server1, I added the other systems info in the client setting menu.
SysGrp-- System-- QueueMgr Prefix--
- FMCGRP FMCSYS O5UAT01A FMC
- FMCGRP FMCSYS1 O5UAT02A FMC
|
You don't have to do this (I mean adding the connect names) on the server.
You will have to do that on the client machines though. Your connect names are also not right. Here is a sample that you should put on the client box.
| Code: |
On the Client1 side client setting menu looks like this
- FMCGRP FMCSYS O5UAT01C FMC
- FMCGRP FMCSYS1 O5UAT01C FMC
|
Notice that the QMGR name is the name of the QMGR the client is residing on. It should be same for both the connect names.
Dont't worry about channel tabs if the communication is all QMGR to QMGR
I tried this setup and it works. |
|
| Back to top |
|
|
| hos |
| Posted: Tue Nov 02, 2004 1:36 am Post subject: |
|
|
Chevalier
Joined: 03 Feb 2002
Posts: 470
|
|
| Back to top |
|
|
| MaheshPN |
| Posted: Tue Nov 02, 2004 11:14 am Post subject: |
|
|
Master
Joined: 21 May 2003
Posts: 245
Location: Charlotte, NC
|
Thanks Vennela,
I configured the way you suggested.
I have couple of questions about the new setup.
1. Now, we have sys ans sys1 names in the client configuration.
Does MQ cluster will take care of load balancing? or It will always send it to sys as it is available?
2. I was trying the failover with the new setup. It works great when I bring down any WF server(SYS/SYS1) with its Queue manager.
But, If I bring down only the WF server, while keeping the Queue manager running, application will try connecting to connect to SYS thinking that it is available. It make sense from MQ perspective as MQ Cluster does not know, whether WF server is available or not. it Just stack up the login request in queues. Do you think the same? Let me know where I went wrong.
thanks,
-Mahesh |
|
| Back to top |
|
|
| vennela |
| Posted: Wed Nov 03, 2004 6:28 am Post subject: |
|
|
Jedi Knight
Joined: 11 Aug 2002
Posts: 4055
Location: Hyderabad, India
|
| Quote: |
1. Now, we have sys ans sys1 names in the client configuration.
Does MQ cluster will take care of load balancing? |
Yes
or It will always send it to sys as it is available?
NO
| Quote: |
| But, If I bring down only the WF server, while keeping the Queue manager running, application will try connecting to connect to SYS thinking that it is available. It make sense from MQ perspective as MQ Cluster does not know, whether WF server is available or not. it Just stack up the login request in queues. Do you think the same? |
YES
That is one another reason to use OS failover capabilities.
One solution would be to check if the workflow server is available and if not suspend the QMGR from the cluster. |
|
| Back to top |
|
|
|
|
