| Author |
Message
|
| olan022 |
 Posted: Tue Oct 29, 2002 11:07 am Post subject: Memory leak on MQ 5.2 CSD04 |
 |
|
Acolyte
Joined: 17 Apr 2002
Posts: 50
|
I am running on AIX 5L with a single 375Mhz and 2GB RAM and see the following situation constantly:
I am noticing the following processes when I start the Queue Manager up.
amqzlaa0 -mQMR -fip1
amqzlaa0 -mQMR -fip2
About 10 or so to begin with.
In a matter of 3 days the number can grow to about 500 or so, give or take. The server memory usage shows 100% on the AIX server and it can lead to the QMGR going out to lunch (sometimes it will come back) due to what seems a lack of resources. Restarting the Queue Manager does not release memory. I tend to think it is a memory leak but wanted to poll the community to see if they perhaps have seen this before. I have seen a post close to this but it was specific to Windows 2000 and didn't know if this would also apply to AIX.
Thanks in advance. |
|
| Back to top |
|
 |
| kp |
| Posted: Tue Oct 29, 2002 11:52 am Post subject: |
|
|
Novice
Joined: 16 Oct 2002
Posts: 15
Location: Belgium
|
I've had the same kind of problem with SUN MQ5.2 but with amqcrsta processes . Applying CSD04 for SUN solved our problem. The only thing I can advise is change CSD.
Note that it is a good idea to clean up your shared memory/message queues/semaphores identifiers on the AIX with ipcrm before you start the qmgr again. It wasn't as clean as we hoped.
Good luck,
kp |
|
| Back to top |
|
|
| olan022 |
| Posted: Tue Oct 29, 2002 12:12 pm Post subject: |
|
|
Acolyte
Joined: 17 Apr 2002
Posts: 50
|
| Anyone see this in AIX-land or know anything more about the amqlzaa0 process? |
|
| Back to top |
|
|
| jmac |
| Posted: Tue Oct 29, 2002 12:59 pm Post subject: |
|
|
Jedi Knight
Joined: 27 Jun 2001
Posts: 3081
Location: EmeriCon, LLC
|
Based on some research I did a year or so ago I found that the amqzlaa0 tasks are for queue manager agents. Now of course, I have no idea what a queue manager agent does, but I did dig this information out of some IBM manual.
Hope this helps
_________________
John McDonald
RETIRED |
|
| Back to top |
|
|
| leongor |
| Posted: Wed Oct 30, 2002 1:31 am Post subject: |
|
|
Master
Joined: 13 May 2002
Posts: 264
Location: Israel
|
Well, when you work with MQCONN (standard connection), yuor program use interprocess communication connecting with qmgr agent.
Every qmgr agent can handle some number of connections ( I think the default value is 52 ). If you make more then new agent arises.
If you're sure that you don't have a lot of connecting applications then I would recommend you to recreate qmgr.
Installing CSD05 may help also.
_________________
Regards.
Leonid.
IBM Certified MQSeries Specialist. |
|
| Back to top |
|
|
| udaybho |
| Posted: Wed Oct 30, 2002 6:08 pm Post subject: |
|
|
Voyager
Joined: 09 May 2002
Posts: 94
Location: Chicago
|
Could you post how's your application is using Work Flow/MQ.
I came across this situtation where I have many application connecting to QM. ANd I found the hanging channels along with Agent Processes.
The solution I applied is by changing TCP: stanza of qm.ini file with keepALive=YES
TCP:
keepAlive=Yes
Do you see many channel processes(amqcrsta) too ?
Uday BHosle |
|
| Back to top |
|
|
| olan022 |
| Posted: Thu Oct 31, 2002 7:53 am Post subject: |
|
|
Acolyte
Joined: 17 Apr 2002
Posts: 50
|
I unfortunately cannot tell you how the applications are being used because there are quite a few out there and different projects are creating their own so it is difficult to diagnose.
About the amqcrsta processes - Now that you mention it, yes.
There are roughly
80 amqzlaa0 (queue manager agent) processes and
60 amqcrsta processes out there.
Doesn't the KEEPALIVE stanza kill stagnant connections after two hours or is that accomplishing something else? Did using that elminate both the amqcrsta and amqzlaa0 processes in your environment from getting too large? |
|
| Back to top |
|
|
| udaybho |
| Posted: Thu Oct 31, 2002 11:12 am Post subject: |
|
|
Voyager
Joined: 09 May 2002
Posts: 94
Location: Chicago
|
KeepAlive=YES lets TCP/IP( I am assuming you are using TCP as your transport) to check the other end of the connection is alive or not. If it is NOT then channel(amqcrsta) is closed.
Now because agent process (amqzlaa0) acts as interface between the client and QManager, their count too decreases as channel count decreases.
You have too many agent process against the channel processes. Usually agent processes should be less then channel processes. You should open a PMR with IBM.
Uday Bhosle |
|
| Back to top |
|
|
| olan022 |
| Posted: Thu Oct 31, 2002 12:32 pm Post subject: |
|
|
Acolyte
Joined: 17 Apr 2002
Posts: 50
|
| I will give that a whirl. Thanks everyone for the advice! |
|
| Back to top |
|
|
|
|
