Integration

Hello,

we are seing problems with RAC and XA (Tuxedo 11, DB 11.2), specifically encountering "ORA-24798: cannot resume the distributed transaction branch on another instance".

The first scenario relates to fallback after a RAC node failure. There are two servers, S1 and S2. S1 makes an ATMI call to S2. Both servers are in the same Tuxedo group, using TMS_ORA. RAC is set up for failover (BASIC), no load balancing.

The sequence is:
- S1 and S2 are connected to the same RAC node n1. All is well.
- RAC node n1 fails. S1, S2 and the TMS_ORA all fail over to RAC node n2. After the failover has happened, all is well.
- RAC node n1 recovers. All is still well (as there is no automatic fallback).
- S1 (or S2) is restarted (either intentionally or because of a crash). Since n1 is up again, S1 connects to n1. Now we get ORA-24798. Permanently.

S1 is connected to n1 and S2 is connected to n2. Since both are in the same group, both use the same XA transaction branch. When called, S2 attempts to JOIN the transaction branch that S1 started. But the DB (11.2) does not allow the same branch to span more than one node. Hence the ORA-24798.

This seems to be a severe limitation in the combination of Tuxedo, XA and RAC. It basically means we still have to use DTP services, even with Tuxedo 11 and DB 11.2. Or are we missing something?

We could put S1 and S2 into different groups, but that seems to be inefficient, and not practical for a real application (10s of servers).

I am extrapolating from this that RAC load balancing would also not work, as S1 and S2 could be connected to different RAC nodes.

Roger