CPU pegged for no apparent reason...

[d3e794]

Hello,

I'm running Mercator version 6.7.1 on Solaris. In general it is running fine, and has been for awhile. Sometimes, however, the eventserver daemon is getting pegged at 100% of a given CPU when there are no files even being processed. When it's in this state, other files can come in and be processed because they use the second CPU on the Unix box, but the first CPU stays pegged until I restart the eventserver entirely. I'm running into this on a system in which I can control incoming files so I know that there is nothing that it SHOULD be processing to get into this state.

Has anyone else had this problem? Any suggestions on tracking it down or fixing it?

Any info would be much appreciated!!

Thanks!
Tim

 

[OllyC]

Possibly one of the threads in the Event Server process has got stuck in a loop - this would cause one CPU to max out while the other is idle.

I think you need to raise it with IBM support to get a resolution though. (they'll probably ask you to run the 'ui' script, you may need the dbx debugger installed too).

 

[eyetry]

I don't thnk they support 6.7.1 anymore but you should be able to do some research on it in their Dev Works area.

 

[JEMIAZ]

6.7.1 is still supported - assuming mainternance support has been paid - as no end of life date has been published yet. We thought it was to be eol in April 08 but that has never been confirmed

 

[janhes]

Are you using any adapters other than file?

 

[d3e794]

Thanks everyone who has already responded! In answer to the adaptors question, I am using Oracle 9i on a second Solaris box. I have checked the database connections, however, and the eventserver/launcher is not even connected to this database at the time of the runaway process.

Also, the runaway CPU has been happening every night for at least three nights now. I turned on some monitoring scripts to see when it is starting, and it appears to start even if no incoming files are processing AT ALL. I'm wondering if perhaps an outside thing such as a backup or system scan could put it into such a state?

Any other ideas?

Also, apologize for the stupid question, but eyetry mentioned "Dev Works area". Where is that at?

Thanks,
Tim

 

[maltu]

Are you using commerce Manager, is there a export map running every hour or so all through the day?

 

[d3e794]

Yes, there is an export happening every hour. My current thought is the last Solaris 9 patch cluster from February is the culprit. I was not having this problem before the Solaris patch cluster was installed, and now I'm consistenly having it with EXACTLY the same code. In fact, I'm using a tar file that contains an exact image of what was running fine before, and is also running fine in the field without the latest patch cluster. The only thing I have left to do is remove the patch cluster here to verify that the problem goes away.

Thanks again to everyone that responded!!!! I'll let you know what happens when we remove the patch cluster.

 

[eyetry]

While 6.7.1 was/is 9i compat in windows you needed to make sure you dealt with the proper dlls. A shot in the dark, but are you pointing to the right files?

- Beginning with this release, the following is supported: the
Oracle 9i adapter (32-bit version) with a 32-bit
Oracle 9i client (Solaris and Windows) up to a 64-bit Oracle
Server (any 64-bit platform that Oracle supports).

- [TR 16909] There are three versions of the Oracle adapter: 8.0,
8i and 9i. The version corresponding to the installed client
should be used.

UNIX
----
The installation program will try to determine which version of
the Oracle adapter you have installed and update it with the
latest revision. If it cannot determine which Oracle adapter is
installed then you will be prompted to choose one.

After the installation is complete, you can always change the
Oracle adapter you are using by following these steps:

1. Change directory to install_dir/libs.
2. Delete the libm4ora8.so (or libm4ora8.sl) file.
3. Copy the desired version of the adapter to replace the file
libm4ora8.so(.sl) as listed here:
Oracle 8.0: libm4ora80.so (.sl) --> libm4ora8.so(.sl)
Oracle 8i: libm4ora8i.so(.sl) --> libm4ora8.so(.sl)
Oracle 9i: libm4ora9i.so(.sl) --> libm4ora8.so(.sl)


SUN SOLARIS MERCATOR PRODUCT REQUIREMENTS
-----------------------------------------

Mercator products that run on Sun Solaris platforms require:

- Solaris versions 2.6 (6), 7, 8, or 9

- Solaris 2.6 (6) must be at Base Kernel/OS level 105181-20 (available in the
recommended Solaris 2.6 Patch Cluster from SunSolve.com via the Sun.com web
site).

- To resolve a failure to listen on TCP sockets, install patches:
105529-03 /kernel/drv/tcp patch
105214-01 /kernel/fs/sockfs patch (available to Contract
Customers only)
Or install the Patch Cluster.

- The following components are supported on Sun Solaris 9:
Command Server
Event Server
Platform API
Java API
EJB API
Servlet Integrator
Mercator Programming Interface
Management Tools (Management Console and Event Server Monitor)
Resource Registry
Event Server Administration

Not supported on Sun Solaris 2.5:
Java Class
JMS