Resource manager does not seem to support failover

Question

The resource manager, TMS_ORACLE, does not seem to support SQLnet's transparent
application failover (TAFO). The way this supposed to work is that if the DB connection
is lost, SQLnet detects it, and establishes a new connection in accordance with
the 'failover_mode' settings in your 'tnsnames.ora' file. It works fine for Oracle
tools like sqlplus, and also for the Tuxedo servers we created. However, the resource
manager doesn't seem to detect the loss at all, which renders our application
useless. Anybody seen this one?

Answer 1

Marcin Kasperski <[email protected]> wrote:
"Tom Hogan" <[email protected]> writes:
The resource manager, TMS_ORACLE, does not seem to support SQLnet's
transparent application failover (TAFO). The way this supposed to
work is that if the DB connection is lost, SQLnet detects it, and
establishes a new connection in accordance with the 'failover_mode'
settings in your 'tnsnames.ora' file. It works fine for Oracle tools
like sqlplus, and also for the Tuxedo servers we created. However,
the resource manager doesn't seem to detect the loss at all, which
renders our application useless. Anybody seen this one?We recently observed something similar - 'normal' Oracle applications
(in our case OCI programs) work perfectly after failover, XA
applications go at the same time crazy (mysterious OCI errors with
strange or no description repeated forever).
IIRC we also noted, that XA server is not provided with the callback
which inform about failing over. Probably it is provided to TMS_ORA.
Nevertheless, there is thing which seems promising: non-failover TNS
alias + setting the server to TPEXIT on (most) Oracle errors (so the
server restart on error and then choose the database which works).
But we did not perform any heavy testing, so in case you experiment
with it more, please share your experiences.
By the way: it is still very unclear to me what TMS_ORA does on
database failure. Using Oracle 8.0.* () I observed that TMS_ORA kept
running after database restart, rejecting all the transactions
received by the server (without even notifying it). Using Oracle 8.1.*
it is better (probably thanks to Oracle 'silently reconnect'
functionality), but ... it is Oracle who improved a bit. We even
opened support case asking for the clarification but the answer didn't
helped much :-(
( Marcin Kasperski | Software is not released, it is allowed to escape.
( http://www.mk.w.pl |
( Porady dla twórców serwisów WWW: http://www.mk.w.pl/porady/porady_www
)Thanks for the reply. I have opened cases with both BEA and Oracle. Oracle is
sure the fault is BEA. BEA says the fault is probably Oracle. Both cases are still
open, so there may be some hope. If I find anything I will post it.
Regards
Tom Hogan