PRIMECLUSTER Global File Services Configuration and Administration Guide 4.2 (Solaris(TM) Operating Environment)

15.2.2 Node layout

Up to four nodes can simultaneously share one GFS file system, therefore, file system processing can continue even if the MDS node goes down. To set up for automatic recovery of the MDS, select two or more nodes on which MDS can operate, then set these nodes as MDS nodes. From the MDS nodes, select the primary MDS node (on which the usual primary MDS operates), the secondary MDS node (on which the usual secondary MDS operates, and all nodes as standby MDS nodes (backup MDS nodes if primary and secondary MDS nodes are unavailable).

For details on the effects of MDS node settings, see Section 14.1, "MDS Failure Recovery."

As previously described in Section 15.2.1, "Effects of file system operation and system load," the following should be taken into consideration:

• CPU and I/O loads accompanying updating of the meta-data of the file system that occur on the node where MDS operates
• CPU throughput, I/O throughput, memory capacity, and swap area

MDS receives requests from all nodes sharing a file system and then updates the meta-data, therefore, install at least two CPUs on nodes on which MDS may operate.

To activate a node that accesses the GFS Shared File System, the node must have been set as an MDS node, or the primary MDS node, secondary MDS node, or standby MDS node (or all together) must be activated simultaneously, or activated in advance. To stop all nodes set as MDS nodes, all nodes that access the relevant file system must have been stopped.

By setting all nodes that share a file system as MDS nodes (i.e., primary MDS node, secondary MDS node or standby MDS node), each node can be activated or stopped regardless of the state of the other nodes. If both the primary and secondary MDS nodes are stopped during operation, the file system can still be used because standby MDS nodes will be activated as the primary MDS and secondary MDS nodes.

When activating the primary or secondary MDS on a standby MDS node, activation of the primary MDS or secondary MDS may take some time.

When MDS is downed, as a result of down recovery processing, it may be in the layout where the primary MDS is not running on the primary server or the secondary MDS is not running on the secondary server.

When returning this MDS layout (where the primary MDS is running on the primary server or the secondary MDS is running on the secondary server), it is able to return by MDS failback with sfcswitch command.

For more information about MDS failback, see Section 14.1.4, "MDS failback."