Forum Discussion

tanislavm's avatar
tanislavm
Level 6
11 years ago

login

Hi,

If an group will failover enough fast then the users not need to relogin to use the application.right?

Only if the failover takes long time,the users should relogin.

 

 

On parallel service group there is no floating IP.right.

The chain is like:user->application->oraclerac.But between application and ORAC there is an software balancer.right?So some transactions that come from the application go through an oracle instance and other transactions go through other oracle instance from ORAC.right?

So if an box with an orac instance fails,the transactions go through the other box.

 

 

 

 

thanks so much.

  • If an group will failover enough fast then the users not need to relogin to use the application.right?

    Only if the failover takes long time,the users should relogin.

    >>> Incorrect, if application is a failover application & configured in cluster, application will be completely offline before it is onlined on second node so users may need to relogin. Time is not a factor here.As application will be offline & online again, session termination is likely to happen.

     

    On parallel service group there is no floating IP.right.

    >>> Correct,parallel service groups will have individual IPs, for e.g oracle RAC database will have IPs for listener on each node where for database to connect to parallel instances.

    The chain is like:user->application->oraclerac.But between application and ORAC there is an software balancer.right?So some transactions that come from the application go through an oracle instance and other transactions go through other oracle instance from ORAC.right?

    >>> Correct, RAC will take care of load balancing between two instances

    So if an box with an orac instance fails,the transactions go through the other box.

    >>> Ideally yes, if I remember correctly, you need to configure TAF (transparent application failover) in TNS to enable seamless movement, otherwise users may need to reattempt connection

     

    G

  • These all depends on the application - below are some scenarios for an application where you are entering some data , when a service group fails over 

    1. The application client has held data you were entering locally and reconnects when the group fails over and the failover is seamless to the client.
       
    2. The application looses the data you were entering, but reconnects when the group fails over so you don't have to reconnect, but have to reenter the data
       
    3. You have to reconnect and have to reenter the data

    The above results may vary depending on how long it takes to failover, compared with the timeouts used in the application - i.e if the application has a timeout of 3 mins to reconnect if it looses connection, then obviously VCS need to failover group in less than 3 mins (including detection time).  Also it may depend on what the application is doing - if the group fails over just as application client sends data that has being inputted, then the application client code MAY resend data if "send" doesn't work or it may not.

    So it all depends on how the application is coded and configured and the same is true of Oracle RAC - it depends on how you configure the client.

    Mike

  • This is more an Oracle question than VCS - my understanding is most of the configuation is done in the oracle client (i.e the client load balanced) - I know of at least 3 methods (but I am not an Oracle DB expert so this maybe incorrect):

    1. Oracle client is configured with one IP, so needs IP to failover on server before it can access Oracle RAC on second node
       
    2. Oracle client is configured with 2 IPs so that when it looses connection to first node, it tries the next IP in its configuration to connect to 2nd node (I would guess the Oracle client can also be configured to send transactions round robin between the 2 IPs)
       
    3. Oracle client is configured with 2 IPs and uses TAF PRECONNECT and sends transaction to BOTH nodes - the primary instance and backup instance, and only the primary instance applies the transaction, but if the primary dies, then the back instance has the transaction, so no transactions are lost (I believe in the other 2 scenarios that Oracle RAC as to roll back any uncompleted transactions).

    As I say, I know Symantec products, not Oracle, so above may be wrong.

    Mike

  • Some sample configurations can be found here, this should clear your doubt on dependency

    https://sort.symantec.com/public/documents/sfha/6.1/solaris/productguides/html/vcs_oracle_agent/apbs01.htm

    so for every parallel instance i need an software balancer.right?

    >> Please read through previous response provided, Once you install RAC, load balance feature will come automatically, its not that you need to install something separtely on each node

     

    G

  • Also look at

    https://sort.symantec.com/public/documents/sfha/6.1/solaris/productguides/html/vcs_oracle_agent/apbs02.htm

    This depicts graphical representation of dependency


    G

  • Just to clarify:

    Your application client will communicate through a virtual IP to an application process on the server and this process will have associated memory about the connection state, so when the application process fails over to another system, the application process is now a different process available on the same virtual IP and it will most likely not have the same in-memory state about the connection, so the application client needs to have the logic to:

    1. Wait for the process to become available again (while VCS fails it over)
    2. Reconnect to the NEW process

    Note this is not the same as if your application client lost access due to network outage, as in this scenario, it would need to do step 1 above "Wait for the process to become available again" while network recovers, but it would probably not need to do 2, as the process still exists (unless the application server process timed out and dropped connection).

    Mike

  • As before it depends on what the client and application are, how they are configured and in some instances how long the failover time is.  i.e in for some clients and applications it won't matter how quick the failover is, the client will need to reconnect, but for other applications the client and/or application server component will have timeouts so the failover time matters.

    Mike