The infamous ORA-28576 "lost RPC connection" error

From: Foo Man Choo SE <foomanchoose_at_yahoo.com>
Date: 8 Oct 2001 15:21:42 -0700
Message-ID: <424fd2be.0110081421.32986b@posting.google.com>

I'm seriously thinking this may be an Oracle bug. We are running Oracle 8.0.4.4 and OAS 4.0.7.1 on Solaris 6 (SPARC) and are having *intermittent* failures when attempting to invoke external routines (C shared lib). The following is an example of the error:

OWS-05101: Execution failed due to Oracle error 28576
ORA-28576: lost RPC connection to external procedure agent
ORA-06512: at "PMQUERY.INFORM", line 0
ORA-06512: at "PMQUERY.PROC_ONSUBMITAXINFORM", line 125
ORA-06512: at line 5

PL/SQL Cartridge SERVICE: pdms/dev
PROCEDURE: proc_onsubmitaxinform

PARAMETERS:

  HTECNUM:
   EC0009297
  HTREVNUM:
   EC0009297-RVW0001
  TUSER:
   user
  TWORKGROUP:

We have verified the following:
(1) The external routine is not crashing when we get this error. The
first thing it does is log its invocation and parameters. If it were to crash, it catches most of the common signals (like segv) and attempts to return gracefully. It has been extensively tested outside of the Oracle environment.
(2) The external routine does not appear to be getting invoked at all
when we get this error. We know this because it hasn't crashed and it hasn't logged its invocation.
(3) The LISTENER.ORA/TNSNAMES.ORA files are configured based on the
defaults as provided by Oracle.
(4) Again, the failures are intermittent. It almost seems as if every
other call is failing. For example, we have an HTML form which results in a post to a handler stored procedure which invokes the external routine. It seems to work the very first time we get on with the browser. However, when we return to the form page (e.g. using the browser back button), it usually fails about every other time we try it (weird!). At first I thought it might be some sort of timing issue, but it seems to follow the every other time pattern even if a few minutes have passed.
(5) In the above scenario, the stored procedure is passing the same
data to the external proc whether it works or fails. We have verfied this directly.

Do you Oracle experts have any insights on this? BTW, I've searched Usenet and Metalink without success. Most cases appear to be related to the extproc crashing.

An aside, WHY-OH-WHY doesn't Oracle provide better, more specific error messages? Surely, it knows which part of the RPC connectivity failed!

--Rob Received on Mon Oct 08 2001 - 17:21:42 CDT