2 replies to this topic

[Dan Shannon]

A client has just gone live and is experiencing regular 'database connection dropped' errors. This appears to be linked to an ORA-03113 error, which is resulting from a fatal error 12571 (according to the Oracle trace files).

Has anyone out there experienced similar things? Is there a way around it? Is it Oracle (it looks like a TNS packet write error) or is it that the PRO-IV kernel has dropped the ball somewhere?

Any (constructive ) advice appreciated

Cheers

Dan Shannon

[Doug Smith]

Hi

Oracle have a document suggesting several fixes for intermittent SQL*NET problem on NT. The document is copyrighted so I can't post it here. It mentions synchronisation errors in the TCP/IP layer on NT as being a possible cause of the problems and provides details of fixes. If you have a an Oracle Support account on the internet, search for document 71921.1. Its titled 'ORA-12151, ORA-12571 errors on Windows NT'

Hope this helps

Doug

[Ralph Gadsby]

A workaround for this problem is to not use the TCP/IP layer for SQL*Net communication. Named Pipes work with ProIV, however you do not get the waiting for database response or record locks processed if you use this workaround.

I have tried all the options that Oracle suggest in the aformentioned document and none of them help with this problem. The problem seems to be Oracle incorrectly reporting back an error from Windows. Oracle is reporting a lost connection when there are 2 blocking operations in progress on a thread. We are working on a kernel level workaround for this so that the record lock and waiting for database response messages work correctly.