I found a really odd config on a server today.
I found a situation where there's a 2 node RAC cluster built.
A new database was created on node 2 as a standard instance.
When whoever built the TNSNAMES.ORA file built it, the HOST part of the TNSNAMES entry used the SCAN host name instead of the host name of node 2.
The service of the Application on the app server kept stopping. The vendor says that application service is finicky and even a fraction of a second of DB disconnection will cause the app service to fall over.
I would assume that if you've setup a tnsnames alias using the RAC SCAN alias that if the engine decided to point you to node 1 and there's no node 1 configured that it might generate a brief database "not there" before falling back to node 2.
I'm just running here on a gut instinct that this could be the problem, wondering if anyone has any real life experience to back this thought up ?