Top
PRIMECLUSTER  Installation and Administration Guide4.3

7.5.1 Notes on Switching a Cluster Application Forcibly

When Forced switch request (Forced startup) of an application is issued, RMS overrides all safety checks and starts the application. So if shared resources which require exclusive control between nodes become Online on the multiple nodes simultaneously, it could result in data corruption or other inconsistencies.

Note

RMS may kill the node on which RMS is not running before starting the application to reduce the risk of data corruption when the Forced startup of an application is issued.

To perform forced startup of an application safely, check whether RMS is running on all nodes in the cluster before starting forced startup according to the following procedure, and if there are the nodes on which RMS is not running, then shut down the nodes.

  1. Check the node state by one of the following methods:

    • Execute the cftool -n command on all nodes.

      fuji2# cftool -n
      Node  Number State       Os      Cpu
      fuji2 1      UP          Linux   Pentium
      fuji3 2      UP          Linux   Pentium
    • Check the CF tree of the Cluster Admin.

  2. . Check the following contents for the node states, and take corrective actions if necessary:

    • Check the node states are all UP.

    • If a LEFTCLUSTER node exists, recover CF from the LEFTCLUSTER state.

      For details, see "PRIMECLUSTER Cluster Foundation Configuration and Administration."

    • If a node with DOWN or UNKNOWN exists, or if a node for which the state is not displayed exists, check whether the operating system of the node has stopped. If the operating system is running, shut down the operating system or restart the system in single-user mode.

  3. Check whether some nodes on which RMS is not running exist among the nodes on which cluster applications will be forcibly started by one of the following methods:

    • Execute the hvdisp -a command on nodes where cluster applications will be started and check that the state of objects whose Type is SysNode is Online.

      fuji2# hvdisp -a
      
      Local System:  fuji2RMS
      Configuration: /opt/SMAW/SMAWRrms/build/config.us
      
      Resource            Type    HostName            State        StateDetails
      -----------------------------------------------------------------------------
      fuji3RMS            SysNode                     Online
      fuji2RMS            SysNode                     Online
      app2                userApp                     Offline
      Machine001_app2     andOp   fuji3RMS
      Machine000_app2     andOp   fuji2RMS            Offline
      ManageProgram000_Cmd_APP2 gRes                        Offline
      app1                userApp                     Offline
      Machine001_app1     andOp   fuji3RMS
      Machine000_app1     andOp   fuji2RMS            Offline
      ManageProgram000_Cmd_APP1 gRes                        Offline
    • Check that the states of all SysNode displayed in the RMS tree of the Cluster Admin are Online.

  4. If nodes which satisfy the following conditions exist, shut down the operating system of the nodes, or restart the system in single-user mode.

    • The node state is UP, and

    • The state of SysNode is not Online.

  5. Execute the Forced switch (hvswitch -f) to forcibly start the application.