| Author |
Message
|
| tricky_knight |
 Posted: Tue Mar 29, 2005 3:35 pm Post subject: resources/listenters running out |
 |
|
Apprentice
Joined: 12 Mar 2005
Posts: 34
|
IF you create a qmgr in Linux. MQ5.3/csd9, and start a listener, why do 5 different listener instances appear when i ps ef | grep run ?
I wonder if this is causing resource issues on my box since I create 6 qmgrs but only five will start at a time. (I must end one to start the other)
I know there was some sort of limitation on Win32, but for Linux? Is there anything I need to look at/tweak?
Seems like there should only be one listener running but I see losts of channel initiators etc for EACH qmgr..
is this normal? |
|
| Back to top |
|
 |
| csmith28 |
| Posted: Tue Mar 29, 2005 4:22 pm Post subject: |
|
|
Grand Master
Joined: 15 Jul 2003
Posts: 1196
Location: Arizona
|
Do you really need six MQManagers?
On all Unix platforms there are shared memory resources that are limited unless you tweak the Kernel. 5 MQManagers seems to be the upper limit on Linux.
The reason that five runmqlsr processes show up on Linux is somewhat of a Linux quirk. Unlike AIX and Solaris when you run a ps -ef | grep runmqlsr or some such. It doesn't just show the process it shows the process threads which would lead one to believe that there are multiple instances of the same process.
Or so I am told.
I shudder to think what kind of output one would have to wade through when doing a ps -ef | grep amqrmppa on a Linux Server with an MQManager that has hundreds of Client connections.
_________________
Yes, I am an agent of Satan but my duties are largely ceremonial. |
|
| Back to top |
|
|
| hopsala |
| Posted: Tue Mar 29, 2005 10:41 pm Post subject: |
|
|
Guardian
Joined: 24 Sep 2004
Posts: 960
|
yea, i've seen this phenomenon a few times in the past. According to MQ Lit, In some UNIX systems who do not use threads internally, you will many times see one process per connecting application.
Sometimes MQ will raise several such processes in advance, even before anyone connected.
One example is indeed the listener, another is amqzlaa0 which is the LQM agent (the process used to manage the connection from MQI).
These processes are only as heavy-duty as the programs connecting to them, if you send 10 msgs per hour, it's OK, if 10000 msgs per sec, then you should start thinking about load-balancing...
Regardless, I agree with csmith that the main question here, is why on earth you need six qmanagers on one machine (??). I've been an MQ consultant and IBM instructor for 4 years now, and have visited dozens of sites around the country, and have never come across an architecture which necessitated six qm's on one machine.
In fact, the only case in which anyone needed more than one qmgr, is another qmgr for test and maybe one other for "test-production" as they call it. But usually these resided on dedicated test machines.
so, I am curious to hear of the need for six... ? |
|
| Back to top |
|
|
| Anirud |
| Posted: Wed Mar 30, 2005 7:18 am Post subject: |
|
|
Master
Joined: 12 Feb 2004
Posts: 285
Location: Vermont
|
|
| Back to top |
|
|
| bower5932 |
| Posted: Wed Mar 30, 2005 7:19 am Post subject: |
|
|
Jedi Knight
Joined: 27 Aug 2001
Posts: 3023
Location: Dallas, TX, USA
|
| In a "real" environment, I can't imagine your needing 6 qmgrs. However, I will admit to actually having 9 on my machine. However, I never have more than two running at one time. All of them are what I would consider a 'development' or 'test' qmgr. |
|
| Back to top |
|
|
| tricky_knight |
| Posted: Wed Mar 30, 2005 12:47 pm Post subject: |
|
|
Apprentice
Joined: 12 Mar 2005
Posts: 34
|
so in a true "clustered" environment you would have a Repository QMGR, Secondary Repos QMGR and the other 8 QMGRS ALL on different harware, each QMGR(not the rep and sec rep) would then run against one instance of the application server in it's own WAS app server cluster.
so if one of the queuemanagers goes down, the gateway rep automatically knows and then sends the messages to the others without missing a beat? |
|
| Back to top |
|
|
| bower5932 |
| Posted: Wed Mar 30, 2005 12:51 pm Post subject: |
|
|
Jedi Knight
Joined: 27 Aug 2001
Posts: 3023
Location: Dallas, TX, USA
|
| tricky_knight wrote: |
| so if one of the queuemanagers goes down, the gateway rep automatically knows and then sends the messages to the others without missing a beat? |
If a queue manager is unavailable, it won't have any more messages routed to it.
However, not to cloud the issue, you can still get a message 'orphaned' on a queue manager in the cluster that is not available. Because of this, clustering is described as load balancing and not failover. If you search here on failover in the clustering category, you'll get hits that go into details. |
|
| Back to top |
|
|
| kman |
| Posted: Wed Mar 30, 2005 10:56 pm Post subject: |
|
|
Partisan
Joined: 21 Jan 2003
Posts: 309
Location: Kuala Lumpur, Malaysia
|
| Quote: |
| each QMGR(not the rep and sec rep) would then run against one instance of the application server in it's own WAS app server cluster. |
I take it you are saying that your rep and sec rep is not in the WAS cluster! That can't happen. They are all in the same cluster. Or am I missing a point? |
|
| Back to top |
|
|
| fjb_saper |
| Posted: Thu Mar 31, 2005 5:18 am Post subject: |
|
|
Grand High Poobah
Joined: 18 Nov 2003
Posts: 20756
Location: LI,NY
|
WAS cluster and MQ cluster are 2 complete different animals.
Do not confuse them.  |
|
| Back to top |
|
|
|
|
