Transferring the Lock information
If you use an NFS server function, set it so that the Lock information for NFS is succeeded.
If a transfer of the Lock information for NFS is not possible, you can get multiple Lock information for the same file. Therefore, the application that uses fcntl(2) and lockf(3C) cannot execute the exclusive control. And it becomes the factor to cause file destruction.
But the following attention is necessary on using this function. Execute the next attention whether an application uses a lock unidentified.
On the node operation with NFS servers, do not use the NFS client function. Do not implement NFS mount.
A file system in which NFS is shared with the operation of NFS servers must be used only from the NFS client. Do not use a file directly from an application on a node on which NFS servers are operating.
When failover occurs, the following processes are performed:
Restarting nfsd(1M), statd(1M), and lockd(1M) daemons
NFS lock recovery processing
The access from the client is reserved until the NFS lock recovery processing is completed though the state of the resource becomes Online when the start of nfsd, statd and lockd is completed.
Protocol
The NFS Versions supported by PRIMECLUSTER vary according to the operating OS.
For the operation on Oracle Solaris 10, NFS Version 3 is supported.
For the operation on Oracle Solaris 11, NFS Version 4 and 3 are supported.
In addition, available protocols vary according to the NFS Versions.
For NFS Version 3, use UDP (recommended).
For NFS Version 4, use TCP (required).
Note
TCP is also available on NFS Version 3. However, when using TCP to perform switchover and failback consecutively, recovery may take more time than when using UDP.
NFS Version 4 does not use UDP.
