GPF Map?

[clk430]

What Mercator map/component generates the GPF file when a GPF occurs?

The reason I ask is that when we get a GPF error, our systems just hangs and we need to restart it. I coded a simple map that monitors for GPF files and will page us when one occurs, but this was pointless because when we get a GPF, no maps will run, including the one that notifies us when a GPF occurs.

Any ideas on what part of Mercator is responsible for prodicing the GPF error report?

We have 6.7 on win2k.

 

[janhes]

GPFs are usually caused by conflict over resources. eg you have 2 systems reading the same file and they both try to delete it.

 

[BocaBurger]

GPF's are caused by memory corruption. This can be caused by a program, an API call (to a 3rd party program), a device driver, or the OS itself; or a combination of the previous.
The GPF report is produced by the OS, not by the program. Also, if you have drwatson enabled you will also get a log from that too. You might also get a user.dmp file.

Unless you have an easily reproducible test case, in a non-production environment, this is really difficult to debug.

If you were running Linux or UNIX, this would be fairly easy to resolve since those OS's include debuggers and there are other free tools with the OS that are used to determine the source of a problem.



BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!

 

[clk430]

I see. Since the GPF reports were placed in our Mercator 6.7 install folder and when we got them the server crashed, I assumed Mercator was generating them.

I wish we had Unix over windows...

Back the drawing boards..thanks for your help.

 

[janhes]

I've also had gpfs when I've had a lot of errors from maps which is probably memory corruption. eg map that runs every minute accessing a database and the db crashes. Generates lots of errors and the event server eventually GPFs

 

[clk430]

Well, we do have a lot fo 'soft erros' eg source not available and target not available (all FTP gets/puts), but things process just fine. The errors have been possibly attributed to a double trigger problem - Basically a file is taking longer than the trigger time setting to be created and reach end of file. One map triggers on the creation and another on the end of file (time stamp update). The first map completes and the second fails when it tries to delete the file on success, but it is now gone.

Perhaps that could be an indication of how we eventually get GPF's as well. I know 6.7 has known memory leaks.

 

[JEMIAZ]

We have been investigating gpf's as well and suspect oracle db connections. We use a single map to poll multiple oracle db's every minute. Currently it causes the ES to crash every 3 days but we are at a loss so far to find a solution. We are using 6.7 on windows and oracle 8i client.
The gpf reports were very confusing until we decided to ignore them! Typically before the gof we also get unrelated maps failing reporting out of memory yet our server stats show use of < 50mb on an 8gb server and very few fuinction failures.
It is a painful process at times

Tim

 

[BocaBurger]

What version of 6.7 and what Oracle 8i client. Oracel had issues until you got to 8.1.7.4.1 or higher. There is a difference between total memory on a Windows box and the memory available to a process. There is another issue. Every map error or warning uses about 52 bytes of RAM. Enough errors between restarts of the Event Server and you can have what appears to be a memory leak. I had one case where there were 40k errors an hour and it took about 2.3 hours to kill the server. Resolution was to fix the problem causing the errors.


BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!

 

[eyetry]

FYI, I understand that at times TX gives false 'out of memory/resources' errors.

I wonder if running the MS SDK 'rebind' against the dlls in the Mercator6.7 folder.

 

[JEMIAZ]

BocaBurger
We are using Oracle 8.1.7.4.1 and the number of function failures and map errors we get before failure is usually in the low 100's range.
We are working on eliminating the errors but as an integration service we don't always control the behaviour of external departments Oracle instances.
Tim

 

[BocaBurger]

Since you are on 6.7.x, I suspect memory corruption, based on the number of errors before failure. If you can reproduce this in a test environment, you could get debug code to help determine the location of the problem, or, if you can simplify the system to a small test case, it could be set up in Support and Engineering. Unfortunately, Windows does not have the debug tools that other OS' have that would make this easier to solve.
Upgrading to a current version (Oracle too) might help.


BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!

 

[eyetry]

FYI: Sorry about any possible confusion... I keep referring to 'rebind' and it should be 'rebase'

I'm off learning a new tool and was a few steps back from this.

again, sorry about any confusion.