| Author |
Message
|
| PeterPotkay |
 Posted: Fri Apr 19, 2002 11:17 am Post subject: |
 |
|
Poobah
Joined: 15 May 2001
Posts: 7716
|
QM1 talks to QM2 via channel QM1.QM2. QM2 talks to QM3 via channel QM2.QM3. In this scenario, there are no applications on QM2. It is simply a hop, and uses name resolution to automatically forward any messages thru itself. (i.e. there is an XMIT queue called QM3 on QM2)
If the hardware that houses QM2 goes down for the count, I would like the channel QM1.QM2 to automatically connect to QM2's backup, QM2A., so that messages may continue to flow over to QM3 through the hop.
I don't think MQ Clusters will help here (for QM2 and QM2A), and I don't want to mix QM1 or QM3 in any cluster. I suppose QM2's hardware must be clustered somehow with QM2A's hardware, no? If so, how do the channels from QM1 and QM3 know to switch over to QM2A? Or is this pie in the sky?
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
 |
| kolban |
| Posted: Sat Apr 20, 2002 6:13 am Post subject: |
|
|
Grand Master
Joined: 22 May 2001
Posts: 1072
Location: Fort Worth, TX, USA
|
| Peter, this is a great question ... I have often wondered how we would achieve this ... lets hope we get some inovative answers ... I'm dry. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Wed Apr 24, 2002 6:33 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7716
|
Can a channel exit somehow recognize that the other side is down, and then take appropriate action? (change the hostname of the sender channel to the backup, or if the other side is not responding, move the message to another XMIT queue that resolves to the backup channel)
Just thinking out loud here for possible solutions. Don't have much experience with exits, so I don't even know if this is possible.
_________________
Peter Potkay
MQSeries Certified Specialist,Developer
[ This Message was edited by: PeterPotkay on 2002-04-25 04:40 ]
[ This Message was edited by: PeterPotkay on 2002-05-01 07:40 ] |
|
| Back to top |
|
|
| oz1ccg |
| Posted: Thu Apr 25, 2002 1:38 am Post subject: |
|
|
Yatiri
Joined: 10 Feb 2002
Posts: 628
Location: Denmark
|
Hi Peter,
a way could be to define two sender channels one to QM2 and one to QM2A, using the same XMIT queue.
The nature of WebSphere MQ is that it only allows one sender to open the XMITQ, therefore this solution requres that you don't use retry at all..
What happens it when the channel drops, the sender will create a message in the CHANNEL-EVENT-QUEUE, the it's up to your monitor tool to try starting the channel again and if that fail try the backup and so on. I've been working with that solution for the same issue.... and it works 
_________________
Regards, Jørgen
Home of BlockIP2, the last free MQ Security exit ver. 3.00
Cert. on WMQ, WBIMB, SWIFT. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Thu Apr 25, 2002 4:06 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7716
|
What are floating ip addresses? I guess it means if Server2 has the address 1234 and then goes down, Server2A comes up with the same ip address? And if yes, the sender channel from QM1 would automatically catch the reciever on Server2, because as far as the channel is concerned, it IS hostname 1234????
WIN2000 for Server2 and Server2A if it makes a dif...
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| oz1ccg |
| Posted: Thu Apr 25, 2002 5:23 am Post subject: |
|
|
Yatiri
Joined: 10 Feb 2002
Posts: 628
Location: Denmark
|
The senario I mentioned earlier QM2 and QM2A don't have to share any hardware, one of our setups are used for complet failover where a whole site might disapear sep 1...
One of the problems using shared hardware (as I see it), is the failover situation where a channel might go down, and when it comes up again it's out of sequence with it's counterpart. We did that setup on AIX some time ago, and we did fix the sequence numbers automaticly..... not nice but it worked...
The way of letting two channels share a XMIT queue give the ability to ensure that the queuemanagers(QM1 & QM3) a using two sets of sequence number towards QM2 and QM2A...
_________________
Regards, Jørgen
Home of BlockIP2, the last free MQ Security exit ver. 3.00
Cert. on WMQ, WBIMB, SWIFT. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Thu Apr 25, 2002 5:26 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7716
|
Jorgen,
What tool did you use to look for those event messages and then start up the backup channel?
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| oz1ccg |
| Posted: Thu Apr 25, 2002 5:53 am Post subject: |
|
|
Yatiri
Joined: 10 Feb 2002
Posts: 628
Location: Denmark
|
Hi Peter,
The first setup was done using CA-unicenter, I've done the same later using MQ-autooperator(BMC).
But it's quite simple; the major trick is having a task monitoring the SYSTEM.ADMIN.CHANNEL.EVENT queue, and then decode messages and make the right decisions upon those messages.
Lets say your Channel-monitor(CHM) receives a "Channel Stopped (reason 2283)" message:
Then it's up to the CHM to deside if it should start the channel again or start to backup channel.. Maybee I would let the CHM retry the channel one time before failover. The failover might require that more channels we're switched... This could be handled with one or more CHM's
Currently I've been working on a setup using something likewise on a cluster setup, where I've used a "sleepin" qmgr in the cluster and a workload exit to do the switch when the primary failed... Hard stuff to code and get it waterproof and avoid messagelooping, droping and stille having work-load balancing without using to much ressources.
_________________
Regards, Jørgen
Home of BlockIP2, the last free MQ Security exit ver. 3.00
Cert. on WMQ, WBIMB, SWIFT. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Thu Apr 25, 2002 5:58 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7716
|
Yeah that's what I was concerned about to. What if the primary goes down halfway thru a batch interval of lets say 50 and the messages are persistent. The 1st 25 made it to the other side but have not been commited by the MCAs. Now the channel goes down and the backup starts. What happens to the 1st 25 messages?
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| oz1ccg |
| Posted: Thu Apr 25, 2002 6:08 am Post subject: |
|
|
Yatiri
Joined: 10 Feb 2002
Posts: 628
Location: Denmark
|
they are all marked INDOUBT in some manner, the sending queuemanager knows this, and the receiving queue manager also knows this...
So the 25 msgs would stay in this situation until the channels we're switched back.
The two Qgmrs should be able to solve this when switched back.. but sometimes the channel status goes into DOUBT... where manual intervention is requred..
The way this can be solved automaticly (I know that Big Blue says it's not nessesary) was design the application programs to cover duplicates.. msgsid and correlid is allready known with same timestamp.
If we do so we could automaticly backout all IBDOUBT messages (requres waterproof design)
I'm sorry there might be snakes in the waterpole... 
_________________
Regards, Jørgen
Home of BlockIP2, the last free MQ Security exit ver. 3.00
Cert. on WMQ, WBIMB, SWIFT. |
|
| Back to top |
|
|
|
|
