INI file and Mercator stability

[antonkov]

Hi,
We recently upgraded from 6.5 to 6.7+SP and since then are having an issue with Mercator hanging after certain number of connection errors.
More specifically several maps try to extract data from Oracle DB every 3min. Once in a while DB temporarily runs out of available processes and rejects a connection:

(from DB_trace.txt)[sup]
<346-376>: Error in: OCISessionBegin
Message : ORA-00020: maximum number of processes (30) exceeded
<346-376>: Returned status: (-1005) Failed to connect to the database
[/sup]
Dozen of rejects is Ok but eventually Mercator generates several sets of GPF-**.txt files (for General Protection Error) in the installation directory and after some time it hangs. It is just like a time bomb, once number of rejects exceeds X (which floats) the system crashes, doesn’t matter it worked fine before the very last error.
Mercator Support advised to set the IgnoreGPFs parameter in INI to 1 so the local errors/crashes will not affect the whole engine (i.e. rest of the running maps). Testing showed that it doesn’t really help, server/service still hangs, which can be cleared only by reboot.
OS used - Win2K
Oracle Client – 8.1.7
Changing max DB connection isn't a solution as there always can be DB downtime, network abruption, etc.

Wonder if anybody ran into this before or has any thought.
Thanks!

 

[jinkys]

What are your database sttings within the ini file?
IdleDB
SLimDB
HLimDB
KeepDB
MinDB

 

[antonkov]

IdleDB=30
;SLimDB=0
;HLimDB=0
KeepDB=20
;MinDB=0

 

[eyetry]

We had some issue with 6.0 and after playing with various settings settled on this....

IdleDB=240
SLimDB=45
HLimDB=60
KeepDB=60
;MinDB=10

It appears you may not be keeping your connection open long enough. If you aren't the DBA have you spoken with your's to see how your DB environment is configured? As a starting point, I'd recommend mirroring that to some degree.

Remember, based on your current settings you can have an idle connection for a max of 30 seconds. If the connection is idle for more than 20 seconds it will be tested before being used.

eyetry

 

[antonkov]

Well, my question was how to avoid a global EventServer crash after several GPFs. The GPS can be caused by number of reasons, the perfectly tuned DB still may not be reachable because of network problems. Currently Mercator handles rejected connection for about 120rounds (map runs every 3 min) and then hangs.
In production, with tuned parameters and good database the rejects will happens less often, but it only means the server will go down in weeks, not in hours, which still is unacceptable.

 

[pgrivell]

Generally when you have a problem like this you're stuck until Mercator get round to fixing it. You could try downgrading to 6.7 no SP, we tried the SP and found it caused more problems than it solved.

Otherwise you could try to move your database extracts out of the realm of the event server and place them inside a GET/DBLOOKUP within a map and detect errors using the VALID function. This may reduce the number of GPF failures and allow you to exit more gracefully from a map.