6.12.3 File System on the Shared Disk Device

According to the type of file system on the shared disk device, perform the following settings and corrective actions.

See

The type of file system that can be used on the shared disk device varies depending on the OS. For details on the file system and notes on use, see "Linux user guide" of each OS.

6.12.3.1 Corrective Actions for the Forced File System Check

If ext3 or ext4 is used for a file system, the file system might forcibly be checked during Online processing of a switching file system. It is part of the ext3 and ext4 specification that file systems are checked when a certain number of mounting has been executed since the last file system check, or a certain period of time has passed.

When the file systems are forcibly checked along with node startup or failover, timeout occurs due to file system Online processing, and PRIMECLUSTER startup or failover might fail.

It is necessary to prevent the file systems from being checked by executing the following command for all the ext3 and ext4 switching files.

# tune2fs -c0 -i0 <device_name>

Example

# tune2fs -c0 -i0 /dev/sdi1

After executing the above command, check if "Maximum mount count :-1", "Check interval:0"is displayed using the following command:

# tune2fs -l /dev/sdi1
[snip]
Mount count: 10
Maximum mount count: -1
[snip]
Check interval: 0 (<none>)
[snip]

Note

If the forcible file system check is prevented, file systems might corrupt due to failures such as disk errors and kernel bug. These failures cannot be detected through file system logging and journaling. The file system corruption might cause data corruption. To prevent this, execute the "fsck - f" command to enable the file system forcible check during periodic maintenance.

6.12.3.2 Corrective Actions for delayed allocation

If ext4 or xfs is used for a file system, take the following corrective actions for delayed allocation.

See

Ext4 and xfs are used to make the allocation of the disk area more efficient, and to improve the writing performance, using their "Delayed Allocation" feature. As a result of the implementation of "Delayed Allocation", there is a possibility that a part of data is lost by OS panic or power supply interruption of servers, because the sojourn time on the memory of data that should be stored on the disk becomes longer.

For the details of delayed allocation, see Storage Administration Guide of the Red Hat, Inc.

ext4
The delayed allocation can be set disable by specifying nodelalloc for mount option in ext4. Specify the mount option of /etc/fstab.pcl file as follows.
```
#RMS#/dev/sdd2        /mnt/swdsk2          ext4     nodelalloc      0 0
```
xfs
The delayed allocation cannot be set disable when xfs is used. Therefore, in order to prevent a part of data not be lost by OS panic or power supply interruption of servers, the application should immediately issue the fsync() call after writing to guarantee writing in the file system.