4 Replies Latest reply: Oct 15, 2007 8:01 AM by 601585 RSS

    Rac session failover??


      We have 2 rac nodes on rhle4.

      I have a problem with session failover. If i shutdown database on one node (ORACLE_SID=db1) it doesn't move session to another node (ORACLE_SID=db2).

      Any ideas?

      Thank you

        • 1. Re: Rac session failover??
          Paul M.
          How did you configure the failover ? And which type of failover are you expecting ?
          • 2. Re: Rac session failover??
            Should not rac automaticly move session on another node if first one fails?

            • 3. Re: Rac session failover??
              Provided you did set up T(ransparent) A(pplication) F(ailover) in the client tnsnames.ora

              Sybrand Bakker
              Senior Oracle DBA
              • 4. Re: Rac session failover??
                You need to read RAC manuals before you complain about the mal-function of fail-over. :)

                You should configure client-side TAF, which is traditional and easy.
                You should configure server-side TAF, which is supported from recent versions.

                Client-side TAF is configured using TNSNAMES.ora as previous comment says.
                Server-side TAF is configured using commands and DBMS_SERVICE package.

                Another thing you should keep in mind is that select(query) session is transparently failed over, but dml session is not transparent. Your current transaction fails when the instance crashes. To enable automatic dml failover, you need to implement some manual codes. Connection is always failed over when successfully configured.

                Another important thing is that threre are couple of options for connection failover. You should consult Oracle RAC manual.

                Also note that you must be prepared for the capacity of the cluster. Imagine that both two instances are fully loaded and suddely one instance crashes. There is a strong possibility that another surviving instance cannot keep pace with the loads of two instances. I believe some of us have those experiences. You should carefully plan the capacity of the cluster.