ORA-12152: TNS: Unable to send break message

 The new Exadata customer came with a problem "ORA-12152: TNS: Unable to send break message".  (What is the in-band breaking and out-of-band breaking we know from Tanel Poder's post:
https://tanelpoder.com/2008/02/05/oracle-hidden-costs-revealed-part-1/ ).

The customer say: 

"  We're testing upgrade to Oracle 19c from 11.2.0.4. And on 19c we got a problem with interrupting the connection between the application (pl/sql developer, sqlplus) on the client machine and the Oracle database: ORA-12152: TNS: Unable to send break message.

This problem is show stopper to upgrade to 19c!

The connection between the application (pl / sql developer, sqlplus) on the client machine and the process on the Oracle server is interrupted if the client application does not generate traffic for 60 minutes. The application is waiting for a response from the long procedure. After the procedure on the server is actually finished, the application is still waiting for the procedure to complete. When trying to interrupt the connection from the application side, we get ORA-12152: TNS: Unable to send break message (Cause: Unable to send break message).

Network engineers don't see any problems.

Simple test case show this problem:

begin
 dbms_lock.sleep(3590);
end;
/
Finished successully.
 

But

begin
 dbms_lock.sleep(3610);
end;
/
is finished unsuccessfully. 

 "

Thanks to the detailed description, the customer's problem became clear. It is not a In-Band or OOB breaking. It is actually Dead Connection Detection : DCD was enchanced in 12c to reduce detection time.

DCD is mechanism which allow the RDBMS server to check if the client is alive.

This feature is configured on server side using sqlnet.expire_time in sqlnet.ora. The probe packet is sent to the client side every sqlnet.expire_time minutes. If database server have got an error then client is dead and server can close this connection. In pre-12c releases this work was done by NS layer in SQL*Net . 

The 12c mechanism is intended to reduce the detection time and minimize load from RDBMS. This new mechanism is based on the TCP-keepalive property of the socket. With this approach TCP-keepalive probes are sent by OS after the connection has been idle for some time. Because these probes are implemented on the OS level then RDBMS rely on socket state (don't need send its own probes).

But in 12c we still able to use the sqlnet.expire_time.
After the customer have set sqlnet.expire_time=10 the error "ORA-12152: TNS: Unable to send break message" disappeared.